{"as_of":"2026-08-15T00:01:00Z","caps":{"database_statements":6,"inbound":100,"outbound":100},"context_digest":"sha256:7c615d35812cbae63c816c7546042cdccbd8ba4efa8d3a98d65591507c24f5c6","coverage":[{"denominator":0,"lane":"reference_resolution","note":"Typed states for the displayed outbound observations.","records_observed":0,"source":"paper_references, paper_reference_links","source_observed_at":null,"state":"measured"},{"denominator":54,"lane":"standing_notices","note":"One-hop event checks from named stored sources.","records_observed":54,"source":"scholarly_work_events, retraction_status_cache","source_observed_at":"2026-08-14T06:32:32.682623+00:00","state":"measured"},{"denominator":54,"lane":"inbound_itemization","note":"Pith citing papers itemized under the disclosed page cap.","records_observed":54,"source":"paper_references, paper_reference_links","source_observed_at":"2026-08-12T04:26:55.147561Z","state":"measured"},{"denominator":1,"lane":"external_citation_measurements","note":"A source-named dated measurement, never combined with another source.","records_observed":1,"source":"arxiv_reference","source_observed_at":"2026-08-05T02:28:24.338817Z","state":"measured"}],"external_citation_measurements":[{"count":0,"observed_at":"2026-08-05T02:28:24.338817Z","source":"arxiv_reference"}],"inbound":[{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-12T04:26:55.147561Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2412.01440","last_updated":"2025-08-10T03:36:55Z","snapshot_observed_at":"2026-08-14T05:55:03.491816Z","submitted_at":"2024-12-02T12:30:35Z","title":"BadPatch: Diffusion-Based Generation of Physical Adversarial Patches","version":5},"reference_index":55,"source":"pdf_text","source_observed_at":"2026-08-12T04:26:55.147561Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2412.01440"},"observation_digest":"sha256:54bd95820ef1cba38e79c96205062912e954c1c0165451749df8509eeb9185cd","observation_id":"f0b58d28-5e3f-40dc-90c4-c2e05ab0f6a8","resolution":{"observed_at":"2026-08-12T04:26:55.147561Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-11T13:52:06.526634Z","title":"T., Jia, X., Sun, W., Wu, C., Chen, J., Hu, X., Li, Y ., Wang, X., Zhang, N., Tuan, L","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2412.12722","last_updated":"2025-09-04T06:43:22Z","snapshot_observed_at":"2026-08-12T13:55:52.046809Z","submitted_at":"2024-12-17T09:38:58Z","title":"Defending LVLMs Against Vision Attacks through Partial-Perception Supervision","version":2},"reference_index":24,"source":"pdf_text","source_observed_at":"2026-08-11T13:52:06.526634Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2412.12722"},"observation_digest":"sha256:406c4d7772a67f50c7e6591cdc94d1d3a21352ff0406d837c4b18d2416708c27","observation_id":"fd6c90b7-9b6e-4aaf-8fc4-9b34e96372f1","resolution":{"observed_at":"2026-08-11T13:52:06.526634Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T15:42:36.704050Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.13965","last_updated":"2025-05-20T06:05:56Z","snapshot_observed_at":"2026-08-14T19:51:03.387786Z","submitted_at":"2025-05-20T06:05:56Z","title":"CAFES: A Collaborative Multi-Agent Framework for Multi-Granular Multimodal Essay Scoring","version":1},"reference_index":62,"source":"arxiv_source","source_observed_at":"2026-08-07T15:42:36.704050Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.13965"},"observation_digest":"sha256:72eace61bbc42587fe3b0e3f7fac85ad0c894049a3a0d13563e3a6962abbbd97","observation_id":"3943095f-48b1-4ac5-998b-80969f43dabb","resolution":{"observed_at":"2026-08-07T15:42:36.704050Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T15:44:36.085184Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.14015","last_updated":"2025-06-19T10:05:42Z","snapshot_observed_at":"2026-08-14T13:58:44.564276Z","submitted_at":"2025-05-20T07:09:13Z","title":"AUTOLAW: Enhancing Legal Compliance in Large Language Models via Case Law Generation and Jury-Inspired Deliberation","version":2},"reference_index":52,"source":"pdf_text","source_observed_at":"2026-08-07T15:44:36.085184Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.14015"},"observation_digest":"sha256:d866078a6ca2fc5bd462ba7e83e4a89d23bd06eec7cfe7bc04caafd315707b26","observation_id":"6e2583d8-e779-42d7-b879-4e1ab81c1394","resolution":{"observed_at":"2026-08-07T15:44:36.085184Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T15:39:19.903205Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.14406","last_updated":"2025-09-09T05:59:48Z","snapshot_observed_at":"2026-08-14T00:05:55.702661Z","submitted_at":"2025-05-20T14:20:30Z","title":"Pierce the Mists, Greet the Sky: Decipher Knowledge Overshadowing via Knowledge Circuit Analysis","version":4},"reference_index":47,"source":"arxiv_source","source_observed_at":"2026-08-07T15:39:19.903205Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.14406"},"observation_digest":"sha256:cfc63c33152b2ed641a0bc6b753d1b71d55d04a42a58aae1dc63f34d9570bd20","observation_id":"22f9ca5c-4d14-4a4e-901e-7f2a23842c78","resolution":{"observed_at":"2026-08-07T15:39:19.903205Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T15:08:08.453311Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.16234","last_updated":"2025-06-11T02:36:18Z","snapshot_observed_at":"2026-08-13T19:45:00.839614Z","submitted_at":"2025-05-22T05:08:27Z","title":"LIFEBench: Evaluating Length Instruction Following in Large Language Models","version":2},"reference_index":100,"source":"pdf_text","source_observed_at":"2026-08-07T15:08:08.453311Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.16234"},"observation_digest":"sha256:a7a32fce2d6e55891271796c7eddc9529d0e31b1d079c2a600bbbff85306edcb","observation_id":"617e84f1-45a8-486b-bc09-a595734d39da","resolution":{"observed_at":"2026-08-07T15:08:08.453311Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T15:08:12.817031Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.16241","last_updated":"2025-05-26T02:28:07Z","snapshot_observed_at":"2026-08-14T06:52:38.334219Z","submitted_at":"2025-05-22T05:19:42Z","title":"Three Minds, One Legend: Jailbreak Large Reasoning Model with Adaptive Stacked Ciphers","version":3},"reference_index":29,"source":"arxiv_source","source_observed_at":"2026-08-07T15:08:12.817031Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.16241"},"observation_digest":"sha256:1a5c4539e8e1ca29cac6e819ab65d5a2bd186ae34d213705b4bf3fa4cc5cebac","observation_id":"a343fa56-fc0b-4d7a-b500-dfdcc7e4929e","resolution":{"observed_at":"2026-08-07T15:08:12.817031Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T14:44:29.819479Z","title":"Yu, Qingsong Wen, and Yang Liu","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.17716","last_updated":"2025-05-23T10:33:14Z","snapshot_observed_at":"2026-08-14T01:50:17.535729Z","submitted_at":"2025-05-23T10:33:14Z","title":"Get Experience from Practice: LLM Agents with Record & Replay","version":1},"reference_index":77,"source":"pdf_text","source_observed_at":"2026-08-07T14:44:29.819479Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.17716"},"observation_digest":"sha256:923d1f897f9011ebcdfaf9672d75d0b53c4d39cc0a5cff7353bc518212c933f0","observation_id":"8a4eb285-b48c-484c-acb6-a60cdd51e0b2","resolution":{"observed_at":"2026-08-07T14:44:29.819479Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T14:23:11.298851Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.19139","last_updated":"2025-05-25T13:22:10Z","snapshot_observed_at":"2026-08-13T08:07:08.098541Z","submitted_at":"2025-05-25T13:22:10Z","title":"The Eye of Sherlock Holmes: Uncovering User Private Attribute Profiling via Vision-Language Model Agentic Framework","version":1},"reference_index":43,"source":"pdf_text","source_observed_at":"2026-08-07T14:23:11.298851Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.19139"},"observation_digest":"sha256:87a83bd5d52d9976dd2396917bbeab99a27b6cf4ad35c9349f03515532cbe8fe","observation_id":"1bba61fd-e6c0-4919-b723-958c78f657f9","resolution":{"observed_at":"2026-08-07T14:23:11.298851Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T14:15:45.830684Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment,","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.19443","last_updated":"2025-05-26T03:00:21Z","snapshot_observed_at":"2026-08-07T14:11:48.229560Z","submitted_at":"2025-05-26T03:00:21Z","title":"Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI","version":1},"reference_index":136,"source":"pdf_text","source_observed_at":"2026-08-07T14:15:45.830684Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.19443"},"observation_digest":"sha256:08aec31c0864eecd1c29788873786327d4a55755ecad081d04a66f921ff3e344","observation_id":"db2b1394-36e8-4d2d-b652-6185eae5278f","resolution":{"observed_at":"2026-08-07T14:15:45.830684Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T12:51:12.934888Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2505.23873","last_updated":"2025-06-17T06:16:20Z","snapshot_observed_at":"2026-08-12T21:59:04.679922Z","submitted_at":"2025-05-29T12:28:05Z","title":"KGMark: A Diffusion Watermark for Knowledge Graphs","version":2},"reference_index":49,"source":"arxiv_source","source_observed_at":"2026-08-07T12:51:12.934888Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2505.23873"},"observation_digest":"sha256:74f59bae4fce27e4ec934bbcf0f5a6d4af30277aa6fc27c397fe3711f867152d","observation_id":"46b3250a-75ae-414a-957b-0abf54ff053d","resolution":{"observed_at":"2026-08-07T12:51:12.934888Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T12:09:55.935988Z","title":"A comprehensive survey in llm(-agent) full stack safety: Data, training and deployment, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2506.02040","last_updated":"2025-09-14T07:48:51Z","snapshot_observed_at":"2026-08-12T23:17:01.506390Z","submitted_at":"2025-05-31T08:01:11Z","title":"Beyond the Protocol: Unveiling Attack Vectors in the Model Context Protocol (MCP) Ecosystem","version":4},"reference_index":52,"source":"pdf_text","source_observed_at":"2026-08-07T12:09:55.935988Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2506.02040"},"observation_digest":"sha256:14759988065658a93ddfa7167f2a02684967949351ff31042c946cc81028f322","observation_id":"0a5dd95a-dc79-4ec0-9dd7-44a8c315ce6f","resolution":{"observed_at":"2026-08-07T12:09:55.935988Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T05:47:06.486187Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2506.07121","last_updated":"2026-08-01T10:21:24Z","snapshot_observed_at":"2026-08-13T08:59:49.304500Z","submitted_at":"2025-06-08T13:07:41Z","title":"Quality-Diversity Red-Teaming: Automated Generation of High-Quality and Diverse Attackers for Large Language Models","version":2},"reference_index":49,"source":"pdf_text","source_observed_at":"2026-08-07T05:47:06.486187Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2506.07121"},"observation_digest":"sha256:dd3a85ebb9c6783869bb68dc224d7e2ee261ed1f65a46e022245eabfedf5c872","observation_id":"17b4c414-b8f2-4f89-9b36-58b936c31ebc","resolution":{"observed_at":"2026-08-07T05:47:06.486187Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-07T00:32:36.685699Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment.arXiv preprint arXiv:2504.15585, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2506.13666","last_updated":"2025-06-16T16:24:31Z","snapshot_observed_at":"2026-08-07T00:25:36.770856Z","submitted_at":"2025-06-16T16:24:31Z","title":"We Should Identify and Mitigate Third-Party Safety Risks in MCP-Powered Agent Systems","version":1},"reference_index":78,"source":"pdf_text","source_observed_at":"2026-08-07T00:32:36.685699Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2506.13666"},"observation_digest":"sha256:537035ab502471dc302187d9e992bfc155545289751f435db1ccb43e584d5eb1","observation_id":"1cd39cb2-ad4d-4097-9cd9-44ea269ef0ba","resolution":{"observed_at":"2026-08-07T00:32:36.685699Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-06T21:11:18.191017Z","title":"A comprehensive survey in llm (- agent) full stack safety: Data, training and deployment,","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2507.00841","last_updated":"2025-07-01T15:10:00Z","snapshot_observed_at":"2026-08-14T11:59:50.949390Z","submitted_at":"2025-07-01T15:10:00Z","title":"SafeMobile: Chain-level Jailbreak Detection and Automated Evaluation for Multimodal Mobile Agents","version":1},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-08-06T21:11:18.191017Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2507.00841"},"observation_digest":"sha256:5d6cf95066f1bef140b1bf8d17484a944bb3f64990093cfcf0d944aa3f065ad7","observation_id":"b438ebc6-52ae-409a-a544-a4776270ec1a","resolution":{"observed_at":"2026-08-06T21:11:18.191017Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-06T17:43:16.320326Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2507.10156","last_updated":"2025-07-14T11:12:30Z","snapshot_observed_at":"2026-08-13T21:26:50.330130Z","submitted_at":"2025-07-14T11:12:30Z","title":"Introducing the Swiss Food Knowledge Graph: AI for Context-Aware Nutrition Recommendation","version":1},"reference_index":71,"source":"pdf_text","source_observed_at":"2026-08-06T17:43:16.320326Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2507.10156"},"observation_digest":"sha256:24fd877592c3eeb67ff7a8bfa984538bbed7f1c8b6e753cecffa297e77c4f0b8","observation_id":"0562f6cf-ecf1-410b-972a-4bcbb3489d07","resolution":{"observed_at":"2026-08-06T17:43:16.320326Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-06T15:06:48.912765Z","title":"A comprehensive survey in llm(-agent) full stack safety: Data, training and deployment","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2507.16731","last_updated":"2025-07-22T16:13:43Z","snapshot_observed_at":"2026-08-14T09:26:08.438894Z","submitted_at":"2025-07-22T16:13:43Z","title":"Collaborative Inference and Learning between Edge SLMs and Cloud LLMs: A Survey of Algorithms, Execution, and Open Challenges","version":1},"reference_index":283,"source":"pdf_text","source_observed_at":"2026-08-06T15:06:48.912765Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2507.16731"},"observation_digest":"sha256:cbeeb404a8abca28c8d8c18b90a3032c6f4735c67ae5027d136feba0977b6db4","observation_id":"fa829d15-d2e1-4545-80f0-6e13031e4a64","resolution":{"observed_at":"2026-08-06T15:06:48.912765Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-06T12:52:01.469405Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2507.21407","last_updated":"2025-08-30T06:01:56Z","snapshot_observed_at":"2026-08-13T01:59:31.983259Z","submitted_at":"2025-07-29T00:27:12Z","title":"Graph-Augmented Large Language Model Agents: Current Progress and Future Prospects","version":2},"reference_index":53,"source":"arxiv_source","source_observed_at":"2026-08-06T12:52:01.469405Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2507.21407"},"observation_digest":"sha256:6d4d6e56ddc8317731e04dfb46f35ead6ed7586a2d43f1e68cb1c2a4d896c157","observation_id":"f9a1bb5b-5315-4926-bd23-7882987630c2","resolution":{"observed_at":"2026-08-06T12:52:01.469405Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T23:04:08.645670Z","title":"arXiv preprint arXiv:2504.15585","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2508.06087","last_updated":"2025-08-08T07:29:33Z","snapshot_observed_at":"2026-08-13T04:09:15.813600Z","submitted_at":"2025-08-08T07:29:33Z","title":"Adaptive Backtracking for Privacy Protection in Large Language Models","version":1},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-08-05T23:04:08.645670Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2508.06087"},"observation_digest":"sha256:62a23cfd2b9c51543a57e18b147673053876fbef7131171b5b51615052cd900f","observation_id":"f81e63ee-310a-4b40-a780-cc2cec28daae","resolution":{"observed_at":"2026-08-05T23:04:08.645670Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T10:38:58.402887Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2509.03871","last_updated":"2025-09-04T04:12:31Z","snapshot_observed_at":"2026-08-08T03:54:26.940042Z","submitted_at":"2025-09-04T04:12:31Z","title":"A Comprehensive Survey on Trustworthiness in Reasoning with Large Language Models","version":1},"reference_index":4,"source":"pdf_text","source_observed_at":"2026-08-05T10:38:58.402887Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.03871"},"observation_digest":"sha256:da35a0d304b27d492354f1d10d7ae547405a8be466c8a77fa91fbca62ce1a843","observation_id":"8360dd42-bfa8-4b80-8ac4-5808aa6baa3c","resolution":{"observed_at":"2026-08-05T10:38:58.402887Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T10:21:47.149354Z","title":"Yu, Qingsong Wen, and Yang Liu","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2509.04191","last_updated":"2025-09-04T13:13:57Z","snapshot_observed_at":"2026-08-14T21:28:35.592240Z","submitted_at":"2025-09-04T13:13:57Z","title":"KubeGuard: LLM-Assisted Kubernetes Hardening via Configuration Files and Runtime Logs Analysis","version":1},"reference_index":95,"source":"pdf_text","source_observed_at":"2026-08-05T10:21:47.149354Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.04191"},"observation_digest":"sha256:781b8d467625e195e769c7572afee0f6c1c349b34a98799c224ce403c36dcde1","observation_id":"6dcbcab6-57cf-46f6-b4f6-c7ef91578aa5","resolution":{"observed_at":"2026-08-05T10:21:47.149354Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T04:50:32.170105Z","title":"A comprehensive survey in llm (- agent) full stack safety: Data, training and deployment,","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2509.05946","last_updated":"2025-09-07T06:46:03Z","snapshot_observed_at":"2026-08-12T11:27:02.882063Z","submitted_at":"2025-09-07T06:46:03Z","title":"Large Language Models for Next-Generation Wireless Network Management: A Survey and Tutorial","version":1},"reference_index":175,"source":"pdf_text","source_observed_at":"2026-08-05T04:50:32.170105Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.05946"},"observation_digest":"sha256:5055461fd9563ea37f079d6f66b34f7ecc4bee92b532ed84631ce06f54bf6e76","observation_id":"627e0278-77d1-45ca-b4ff-497df50cc0d9","resolution":{"observed_at":"2026-08-05T04:50:32.170105Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-04T14:42:50.674982Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment.arXiv preprint arXiv:2504.15585,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2509.23573","last_updated":"2026-05-28T01:14:19Z","snapshot_observed_at":"2026-08-04T22:20:08.925642Z","submitted_at":"2025-09-28T02:08:27Z","title":"Uncovering Vulnerabilities of LLM-Assisted Cyber Threat Intelligence","version":5},"reference_index":22,"source":"pdf_text","source_observed_at":"2026-08-04T14:42:50.674982Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.23573"},"observation_digest":"sha256:2657ec197784402c39f78047dc2edb38752c1b9e93535c33f5aca2f73a597206","observation_id":"5687e63f-1cfc-491f-8ddc-4cba96ebfa15","resolution":{"observed_at":"2026-08-04T14:42:50.674982Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-04T14:41:50.921051Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment.arXiv preprint arXiv:2504.15585, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2509.24380","last_updated":"2026-07-04T03:50:21Z","snapshot_observed_at":"2026-08-14T09:02:55.806123Z","submitted_at":"2025-09-29T07:29:18Z","title":"Agentic Services Computing","version":3},"reference_index":178,"source":"pdf_text","source_observed_at":"2026-08-04T14:41:50.921051Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.24380"},"observation_digest":"sha256:2ba3a9371aaac16acf5eb9c38304e796836232a9eec20caca68f61e3d7ef50e0","observation_id":"3004583b-014b-4f1f-a1fb-04940e398556","resolution":{"observed_at":"2026-08-04T14:41:50.921051Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2509.26100","last_updated":"2026-05-14T06:48:52Z","snapshot_observed_at":"2026-08-14T10:53:26.197154Z","submitted_at":"2025-09-30T11:20:41Z","title":"AgenticEval: Toward Agentic and Self-Evolving Safety Evaluation of Large Language Models","version":2},"reference_index":12,"source":"pdf_text","source_observed_at":"2026-05-18T12:35:01.443896Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2509.26100"},"observation_digest":"sha256:55e996f864f1a3283d619fbd030cdffd1ad815cdb35e14c514f76049d19b1e15","observation_id":"2fa5d3d6-f2b3-4950-8354-f4c2c0d9357f","resolution":{"observed_at":"2026-05-18T12:36:22.463634Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-04T09:15:21.818144Z","title":"Yu, Qingsong Wen, and Yang Liu","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2510.16492","last_updated":"2026-06-26T11:50:13Z","snapshot_observed_at":"2026-08-14T23:06:26.856958Z","submitted_at":"2025-10-18T13:22:19Z","title":"Check Yourself Before You Wreck Yourself: Selectively Quitting Improves LLM Agent Safety","version":4},"reference_index":15,"source":"arxiv_source","source_observed_at":"2026-08-04T09:15:21.818144Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2510.16492"},"observation_digest":"sha256:415179cd45d1c5e09176a8dd8f136db1e03ebabd93cf75b936e614a660860bd6","observation_id":"0ccd089e-dbac-49e9-896e-a4b7c58c5b3e","resolution":{"observed_at":"2026-08-04T09:15:21.818144Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-04T08:44:56.504049Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment.arXiv preprint arXiv:2504.15585, 2025a","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2510.19420","last_updated":"2026-05-26T15:57:06Z","snapshot_observed_at":"2026-08-08T13:12:40.582936Z","submitted_at":"2025-10-22T09:43:32Z","title":"Securing Multi-Agent Systems Against Corruptions via Node Contribution Backpropagation","version":2},"reference_index":21,"source":"pdf_text","source_observed_at":"2026-08-04T08:44:56.504049Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2510.19420"},"observation_digest":"sha256:1df91b9b245da972959e219ab7257128cbbda8a2edf4b2ac3cd9ff0717b14732","observation_id":"1f81fcd7-473e-4ead-8dca-e4c2f59e4af5","resolution":{"observed_at":"2026-08-04T08:44:56.504049Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2511.12710","last_updated":"2026-05-18T06:50:41Z","snapshot_observed_at":"2026-07-06T22:35:55.936840Z","submitted_at":"2025-11-16T17:52:07Z","title":"Evolve the Method, Not the Prompts: Evolutionary Synthesis of Jailbreak Attacks on LLMs","version":2},"reference_index":47,"source":"pdf_text","source_observed_at":"2026-05-21T18:58:53.183734Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2511.12710"},"observation_digest":"sha256:6f1575541aac19f241606c570e1dab7ea209805e6ba3ed65f4e4bf905bcf8d16","observation_id":"8990c204-c015-4c3a-bd3e-0608db533c45","resolution":{"observed_at":"2026-05-21T19:00:30.397180Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-03T06:23:43.757957Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment.arXiv preprint arXiv:2504.15585,","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2602.02557","last_updated":"2026-05-29T23:48:47Z","snapshot_observed_at":"2026-08-10T01:54:25.942704Z","submitted_at":"2026-01-30T14:23:50Z","title":"The Alignment Curse: Modality Alignment Supercharges Audio Attacks via Text Transfer","version":2},"reference_index":19,"source":"pdf_text","source_observed_at":"2026-08-03T06:23:43.757957Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2602.02557"},"observation_digest":"sha256:ce0518be9424addb36cfcb2630bfc6a37ada0ecc920cc32b7d0631e22d959d1b","observation_id":"432e8ebe-e187-4091-93ad-ae6b1c7facf7","resolution":{"observed_at":"2026-08-03T06:23:43.757957Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-02T23:43:07.687190Z","title":"org/CorpusID:143424870","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2602.12966","last_updated":"2026-06-09T04:02:52Z","snapshot_observed_at":"2026-08-02T23:42:58.142462Z","submitted_at":"2026-02-13T14:33:13Z","title":"ProbeLLM: Automating Principled Diagnosis of LLM Failures","version":2},"reference_index":2019,"source":"pdf_text","source_observed_at":"2026-08-02T23:43:07.687190Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2602.12966"},"observation_digest":"sha256:c6900619cd1f57cb67528083cb7ed5d1d9f2ac9a9fbaf03e01b97d242c68ca04","observation_id":"746a367d-6d75-4b97-850f-e419c3b57eb1","resolution":{"observed_at":"2026-08-02T23:43:07.687190Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2604.11309","last_updated":"2026-04-13T11:12:30Z","snapshot_observed_at":"2026-08-10T14:33:23.383610Z","submitted_at":"2026-04-13T11:12:30Z","title":"The Salami Slicing Threat: Exploiting Cumulative Risks in LLM Systems","version":1},"reference_index":1,"source":"pdf_text","source_observed_at":"2026-05-10T16:07:31.602378Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2604.11309"},"observation_digest":"sha256:a9896b117bdc41095d27af90c8167a43db88f5bedffb59bdcb61e1199f1b174c","observation_id":"4125b05d-325a-465f-9376-34ee1f85469b","resolution":{"observed_at":"2026-05-11T09:16:04.234442Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2604.11934","last_updated":"2026-04-13T18:22:50Z","snapshot_observed_at":"2026-08-11T15:18:07.150130Z","submitted_at":"2026-04-13T18:22:50Z","title":"BiasIG: Benchmarking Multi-dimensional Social Biases in Text-to-Image Models","version":1},"reference_index":17,"source":"pdf_text","source_observed_at":"2026-05-10T15:33:15.025940Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2604.11934"},"observation_digest":"sha256:c82d2e9a9d89a62d9a9fdf0013bf7c729d63e0acd6c57681c4dbbbb3214599d7","observation_id":"e74ea2e4-43ec-46d2-ac23-36da4267756b","resolution":{"observed_at":"2026-05-11T10:21:00.645817Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2604.16321","last_updated":"2026-02-25T07:55:49Z","snapshot_observed_at":"2026-08-11T00:05:01.451986Z","submitted_at":"2026-02-25T07:55:49Z","title":"LLM-Based Multi-Agent Systems for Code Generation: A Multi-Vocal Literature Review","version":1},"reference_index":63,"source":"pdf_text","source_observed_at":"2026-05-15T19:52:49.324500Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2604.16321"},"observation_digest":"sha256:5a089af323403a07ea3d6b135b3d4d2ef0978ce887c71227c7f8ada5d4f87f97","observation_id":"1e41d869-3c91-4f07-b6bc-d6471d446470","resolution":{"observed_at":"2026-05-15T19:56:33.814418Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2604.19083","last_updated":"2026-04-21T04:52:38Z","snapshot_observed_at":"2026-08-11T14:30:07.776382Z","submitted_at":"2026-04-21T04:52:38Z","title":"ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety","version":1},"reference_index":178,"source":"arxiv_source","source_observed_at":"2026-05-10T03:00:34.862711Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2604.19083"},"observation_digest":"sha256:275567949e7af4d8d491d74505360fb09a506666b85d7c9d424ef3e070f4e950","observation_id":"93e9fd91-8f12-4b4d-acc2-50be6e16a428","resolution":{"observed_at":"2026-05-11T12:46:05.697693Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2604.27861","last_updated":"2026-04-30T13:44:01Z","snapshot_observed_at":"2026-08-11T02:06:00.289012Z","submitted_at":"2026-04-30T13:44:01Z","title":"TwinGate: Stateful Defense against Decompositional Jailbreaks in Untraceable Traffic via Asymmetric Contrastive Learning","version":1},"reference_index":27,"source":"pdf_text","source_observed_at":"2026-05-07T06:12:37.845017Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2604.27861"},"observation_digest":"sha256:97f7e29c552f09022c14b37561366a9016b7613d6190f664e8615de0976831f5","observation_id":"9297d37a-a513-4dcd-8b8d-0d9f71da619c","resolution":{"observed_at":"2026-05-12T10:26:29.219146Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.01899","last_updated":"2026-05-03T14:28:08Z","snapshot_observed_at":"2026-08-14T06:23:41.868803Z","submitted_at":"2026-05-03T14:28:08Z","title":"Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety Alignment","version":1},"reference_index":10,"source":"pdf_text","source_observed_at":"2026-05-09T17:24:54.796037Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.01899"},"observation_digest":"sha256:9adcaab15a8d566fbeef11275af2f9e269dc3d6027e0f71167174b1751ff222a","observation_id":"5b14aa25-c446-4733-8b09-f9ad1b3e762e","resolution":{"observed_at":"2026-05-11T16:21:07.319493Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.11679","last_updated":"2026-05-13T09:28:34Z","snapshot_observed_at":"2026-08-11T08:28:39.520767Z","submitted_at":"2026-05-12T07:38:59Z","title":"Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion","version":1},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-05-13T01:03:10.263663Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.11679"},"observation_digest":"sha256:2334c68ecdd207f331ddef6b73b6d5f3bab06af2f0a9dd10c986f9e6a623579b","observation_id":"0536dfdb-3e49-4208-97fc-b63d716f6d09","resolution":{"observed_at":"2026-05-13T01:07:00.563071Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.11679","last_updated":"2026-05-13T09:28:34Z","snapshot_observed_at":"2026-08-11T08:28:39.520767Z","submitted_at":"2026-05-12T07:38:59Z","title":"Explaining and Breaking the Safety-Helpfulness Ceiling via Preference Dimensional Expansion","version":2},"reference_index":7,"source":"pdf_text","source_observed_at":"2026-05-14T21:12:06.989077Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.11679"},"observation_digest":"sha256:4696356a14d54bb6d18023e6a567d86317140c5a2c7fd2c4ff262f32afa5fcac","observation_id":"0eb13ae8-32e2-4dc7-8a0b-27627098488d","resolution":{"observed_at":"2026-05-14T21:12:58.976961Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.12529","last_updated":"2026-04-15T10:56:08Z","snapshot_observed_at":"2026-08-12T22:25:40.940152Z","submitted_at":"2026-04-15T10:56:08Z","title":"BackFlush: Knowledge-Free Backdoor Detection and Elimination with Watermark Preservation in Large Language Models","version":1},"reference_index":9,"source":"pdf_text","source_observed_at":"2026-05-14T21:01:10.756844Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.12529"},"observation_digest":"sha256:c57c635b6b8014973437f63a78e412801adcb8f409b9744d5dd073da1e2f1181","observation_id":"a6dfaaca-478c-4429-bd9a-077cbb8d879f","resolution":{"observed_at":"2026-05-14T21:02:58.891963Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.16282","last_updated":"2026-04-11T04:25:19Z","snapshot_observed_at":"2026-08-12T21:07:09.004725Z","submitted_at":"2026-04-11T04:25:19Z","title":"Taxonomy and Consistency Analysis of Safety Benchmarks for AI Agents","version":1},"reference_index":52,"source":"arxiv_source","source_observed_at":"2026-05-21T01:42:55.693115Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.16282"},"observation_digest":"sha256:439df3a2608e6e5ae00b9da62330f2e9628b912657dfb1dc3de08644d6ba3511","observation_id":"9a3344e2-5e67-4e03-bc5a-0669f6456152","resolution":{"observed_at":"2026-05-21T01:43:56.850450Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.20266","last_updated":"2026-05-18T20:21:32Z","snapshot_observed_at":"2026-08-14T22:20:00.600867Z","submitted_at":"2026-05-18T20:21:32Z","title":"A Survey of Large Audio Language Models: Generalization, Trustworthiness, and Outlook","version":1},"reference_index":21,"source":"pdf_text","source_observed_at":"2026-05-21T07:38:23.099479Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.20266"},"observation_digest":"sha256:8df65c1169659c2e560fe51b2aede135e8dd5750bf9331e1836cffc38df046bf","observation_id":"1ebf339c-e46d-4d65-addf-a2d9d2dc71a7","resolution":{"observed_at":"2026-05-21T07:39:49.124688Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.20641","last_updated":"2026-05-20T02:55:56Z","snapshot_observed_at":"2026-08-14T14:35:04.135894Z","submitted_at":"2026-05-20T02:55:56Z","title":"Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs","version":1},"reference_index":26,"source":"pdf_text","source_observed_at":"2026-05-21T04:45:35.079192Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.20641"},"observation_digest":"sha256:573f085d9b46bd64445ca3472e4a71b71f48894efcf6c6eaa723142750c419a1","observation_id":"49e329d1-315d-45f4-a7b7-b335c0985d17","resolution":{"observed_at":"2026-05-21T04:49:35.654769Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.25603","last_updated":"2026-05-25T08:54:55Z","snapshot_observed_at":"2026-08-10T09:24:29.678831Z","submitted_at":"2026-05-25T08:54:55Z","title":"Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancy","version":1},"reference_index":43,"source":"pdf_text","source_observed_at":"2026-06-29T21:47:17.894881Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.25603"},"observation_digest":"sha256:835b28fc0950d37b207e48c7ab21e3ae8f4d9ca71339b2b2ac2fd75f9c2d996f","observation_id":"063b7fbf-a09c-4ef6-82f4-5e3ceab79311","resolution":{"observed_at":"2026-06-29T21:53:59.521103Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2605.29396","last_updated":"2026-05-28T05:46:38Z","snapshot_observed_at":"2026-08-14T10:19:02.174588Z","submitted_at":"2026-05-28T05:46:38Z","title":"Aligned but Fragile: Enhancing LLM Safety Robustness via Zeroth-Order Optimization","version":1},"reference_index":27,"source":"pdf_text","source_observed_at":"2026-06-29T07:41:03.219581Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2605.29396"},"observation_digest":"sha256:a41ce3982ba008a22b1a7bc581afe84669dbee9deb06e9cc81075ec6056c9839","observation_id":"a1cdd3b8-337d-4c90-95a3-6dd4ab19b69b","resolution":{"observed_at":"2026-06-29T07:43:13.502152Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.02282","last_updated":"2026-06-01T14:05:35Z","snapshot_observed_at":"2026-08-13T15:14:40.884744Z","submitted_at":"2026-06-01T14:05:35Z","title":"POIROT: Interrogating Agents for Failure Detection in Multi-Agent Systems","version":1},"reference_index":13,"source":"pdf_text","source_observed_at":"2026-06-28T14:44:21.487169Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.02282"},"observation_digest":"sha256:ed5e46bf9562647cb8b024f73fe5dfc2c5a5d16a956d15f6369a90c95ff93940","observation_id":"df30c543-00b4-46cf-9de7-8d4e668bdb15","resolution":{"observed_at":"2026-07-01T23:06:19.985654Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.05660","last_updated":"2026-06-04T03:43:09Z","snapshot_observed_at":"2026-08-08T10:35:54.242394Z","submitted_at":"2026-06-04T03:43:09Z","title":"Safe Embodied AI for Long-horizon Tasks: A Cross-layer Analysis of Robotic Manipulation","version":1},"reference_index":11,"source":"arxiv_source","source_observed_at":"2026-06-28T01:46:36.081851Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.05660"},"observation_digest":"sha256:a382bc3902425b909d90e1a4eaedc6f57c1b0073f7b649c1d236a3b603d4ed05","observation_id":"aaa70230-6616-4e55-ad12-b59207d03a12","resolution":{"observed_at":"2026-07-02T12:56:56.715862Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.10749","last_updated":"2026-06-09T12:01:07Z","snapshot_observed_at":"2026-08-14T19:50:49.492797Z","submitted_at":"2026-06-09T12:01:07Z","title":"Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation","version":1},"reference_index":182,"source":"pdf_text","source_observed_at":"2026-06-27T12:55:22.831264Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.10749"},"observation_digest":"sha256:8d2fa9e170770250c97e0e9190d772e200d93296ae9df5863d369622bc93710c","observation_id":"090f64dc-976d-4b53-b52f-a84b53fc51b1","resolution":{"observed_at":"2026-06-27T13:20:56.866720Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.12474","last_updated":"2026-06-10T05:37:58Z","snapshot_observed_at":"2026-08-13T23:02:16.923310Z","submitted_at":"2026-06-10T05:37:58Z","title":"SAIGuard: Communication-State Simulation for Proactive Defense of LLM Multi-Agent Systems","version":1},"reference_index":3,"source":"arxiv_source","source_observed_at":"2026-06-27T08:04:15.004591Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.12474"},"observation_digest":"sha256:b8b740e9e2f0b5f23e1a08b6b2af29c8a94adad589a5bd2c4551a8e3f6a36a7f","observation_id":"7b03d879-f0ea-47f2-8cf2-a175c1c308fc","resolution":{"observed_at":"2026-07-03T13:28:18.834488Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.25442","last_updated":"2026-06-24T06:10:33Z","snapshot_observed_at":"2026-08-12T14:50:59.166106Z","submitted_at":"2026-06-24T06:10:33Z","title":"PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models","version":1},"reference_index":1,"source":"arxiv_source","source_observed_at":"2026-06-25T21:09:19.727723Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.25442"},"observation_digest":"sha256:19fccbf2d3851ef4e342ff5bac1a23bac0c72db966d50c8e2f3f2803e32022ef","observation_id":"ef2b4891-04f4-4c96-a460-7e69a64d9593","resolution":{"observed_at":"2026-07-04T19:40:06.706686Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.26106","last_updated":"2026-05-01T16:22:46Z","snapshot_observed_at":"2026-08-08T12:40:08.409588Z","submitted_at":"2026-05-01T16:22:46Z","title":"Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints","version":1},"reference_index":9,"source":"arxiv_source","source_observed_at":"2026-07-01T07:56:07.948696Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.26106"},"observation_digest":"sha256:4f183abd425446f306dbdf1add0df16bd285b678829a86982962cf438dda1bc1","observation_id":"b1fc6796-572d-4c6b-b90d-f88057d94645","resolution":{"observed_at":"2026-07-01T08:05:31.326711Z","resolver_source":"arxiv_id","status":"metadata_mismatch"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":"2504.15585","doi":"10.48550/arxiv.2504.15585","metadata_source":"arxiv_reference","pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-05T02:28:24.338817Z","title":"A comprehensive survey in llm (-agent) full stack safety: Data, training and deployment","venue":"ArXiv.org","work_id":"890e4d27-50d3-4726-8e34-bbd56d407411","year":2025},"citing_paper":{"arxiv_id":"2606.29239","last_updated":"2026-08-03T15:22:02Z","snapshot_observed_at":"2026-08-13T00:36:05.768917Z","submitted_at":"2026-06-28T07:06:46Z","title":"Breaking the Rounding Trap: Securing LLMs against Quantization-Conditioned Backdoors","version":1},"reference_index":58,"source":"pdf_text","source_observed_at":"2026-06-30T07:47:18.350953Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.29239"},"observation_digest":"sha256:e9115567368ddf99747d81a15af77789945f14a46a4c657bbbf109f71befe1b5","observation_id":"44eabc72-7363-4262-a7ca-ec27128cfdb2","resolution":{"observed_at":"2026-06-30T08:04:28.739091Z","resolver_source":"arxiv_id","status":"verified_exact"},"standing_notice":{"events":[],"observation":"No event found in the named queried sources as of 2026-08-14T06:32:32.682623+00:00.","reason":null,"source_receipts":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"state":"measured"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-04T04:39:06.922277Z","title":null,"venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2606.29239","last_updated":"2026-08-03T15:22:02Z","snapshot_observed_at":"2026-08-13T00:36:05.768917Z","submitted_at":"2026-06-28T07:06:46Z","title":"Breaking the Rounding Trap: Securing LLMs against Quantization-Conditioned Backdoors","version":2},"reference_index":57,"source":"pdf_text","source_observed_at":"2026-08-04T04:39:06.922277Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2606.29239"},"observation_digest":"sha256:771e5cc894078b613caf7b9713b5112ef62b5c1b6bdb3fd0c49e50288f9b62de","observation_id":"baf5f170-7c2e-4782-985c-c1ae06ee600f","resolution":{"observed_at":"2026-08-04T04:39:06.922277Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-07-14T11:21:48.935912Z","title":"A comprehensive survey in LLM(-agent) full stack safety: Data, training and deployment, 2025","venue":null,"work_id":null,"year":2025},"citing_paper":{"arxiv_id":"2607.10487","last_updated":"2026-07-11T21:48:53Z","snapshot_observed_at":"2026-08-14T19:54:50.341441Z","submitted_at":"2026-07-11T21:48:53Z","title":"Temporary Authority, Permanent Effects: Commit-Time Authorization for LLM Agents","version":1},"reference_index":59,"source":"pdf_text","source_observed_at":"2026-07-14T11:21:48.935912Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2607.10487"},"observation_digest":"sha256:31abc91a16d095ab881bad2407f5d1d589e2e64fbacc8764b799b367bc41d2db","observation_id":"6fc0892e-e9c5-4435-a09e-b5039a8bd066","resolution":{"observed_at":"2026-07-14T11:21:48.935912Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}},{"citation":{"cited_paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment","version":4},"cited_work":{"arxiv_id":null,"doi":null,"metadata_source":null,"pith_arxiv_id":"2504.15585","snapshot_observed_at":"2026-08-01T19:02:48.799933Z","title":"arXiv preprint arXiv:2504.15585 , year=","venue":null,"work_id":null,"year":null},"citing_paper":{"arxiv_id":"2607.17117","last_updated":"2026-07-19T08:07:32Z","snapshot_observed_at":"2026-08-14T20:24:43.518178Z","submitted_at":"2026-07-19T08:07:32Z","title":"Persistent Sparse Autoencoders: Learning Feature Timescales in Language Models","version":1},"reference_index":185,"source":"arxiv_source","source_observed_at":"2026-08-01T19:02:48.799933Z"},"links":{"cited_paper":"/paper/2504.15585","citing_paper":"/paper/2607.17117"},"observation_digest":"sha256:0c04b284dde8a4487e42a44a1828d479a6bb2e178a6f017ac354789700adbb2e","observation_id":"9f185a3e-dd8e-4ffa-9144-a44a7613bf20","resolution":{"observed_at":"2026-08-01T19:02:48.799933Z","resolver_source":null,"status":"unresolved"},"standing_notice":{"events":[],"reason":"canonical_work_link_unavailable","source_receipts":[],"state":"unavailable"}}],"links":{"evidence":"/evidence","html":"/paper/2504.15585/citation-record","integrity":"/paper/2504.15585/integrity","json":"/paper/2504.15585/citation-record.json","paper":"/paper/2504.15585"},"outbound":[],"paper":{"arxiv_id":"2504.15585","last_updated":"2025-06-09T02:36:20Z","latest_version":4,"primary_category":"cs.CR","snapshot_observed_at":"2026-08-09T14:46:30.842437Z","submitted_at":"2025-04-22T05:02:49Z","title":"A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment"},"reference_resolution":{"displayed":0,"state_counts":{"malformed_identifier":0,"metadata_mismatch":0,"parse_uncertain":0,"unresolved":0,"verified_exact":0,"verified_fuzzy":0},"total_outbound_references":0},"refusal":"A citation records a reference. It does not transfer a finding from one paper to another.","schema":"pith.paper-citation-record.v1","standing_sources":[{"observed_at":"2026-08-14T06:32:32.682623+00:00","source":"crossref"},{"observed_at":"2026-08-14T06:32:18.44784+00:00","source":"retraction_watch"}],"thesis":"As of 15 August 2026, this Paper Citation Record lists 0 of 0 outbound references and 54 inbound Pith citation observations for arXiv:2504.15585."}