Paper Citation Record · LEDGER

COSMIC: Generalized Refusal Direction Identification in LLM Activations

As of 17 August 2026, this Paper Citation Record lists 23 of 23 outbound references and 3 inbound Pith citation observations for arXiv:2506.00085.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2506.00085 v1

Coverage vector

measured 23 of 23 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-07T12:35:40.431529Z

measured 26 of 26 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-17T06:30:58.91139+00:00

measured 3 of 3 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-06-28T10:04:13.562338Z

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: arxiv_reference, observed 2026-07-04T10:09:45.199419Z

Reference resolution

23 of 23 outbound references displayed

verified exact0
verified fuzzy10
unresolved13
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 19206fc7-6e2c-4b2f-b7ab-9d4eb1f350a3 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 1

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:43.664169Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 77a98af4-5d6a-417e-af97-297c8def4474 · outbound

This paper cites Amrita Bhattacharjee, Shaona Ghosh, Traian Rebedea, and Christopher Parisien.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Amrita Bhattacharjee, Shaona Ghosh, Traian Rebedea, and Christopher Parisien

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:44.750434Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 48aa8dea-541a-41a4-9058-66c768a2dd23 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 3

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:43.280470Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 95c3400d-7db9-4845-8e13-024068cef4b0 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 4

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:43.090508Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 6952dc47-d76a-43fd-ac60-3c9ddd378157 · outbound

This paper cites right to explanation.

COSMIC: Generalized Refusal Direction Identification in LLM Activations right to explanation

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:44.432834Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 5e6d37d3-9821-4d1d-afdd-057ef1955d7d · outbound

This paper cites Mantas Mazeika, Long Phan, Xuwang Yin, Andy Zou, Zifan Wang, Norman Mu, Elham Sakhaee, Nathaniel Li, Steven Basart, Bo Li, David A.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Mantas Mazeika, Long Phan, Xuwang Yin, Andy Zou, Zifan Wang, Norman Mu, Elham Sakhaee, Nathaniel Li, Steven Basart, Bo Li, David A

Reference 6

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:44.268184Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 77e6bcea-72a4-4f40-aedc-3c1047fb9f50 · outbound

This paper cites On Prompt-Driven Safeguarding for Large Language Models.

COSMIC: Generalized Refusal Direction Identification in LLM Activations On Prompt-Driven Safeguarding for Large Language Models

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-07T12:35:39.020367Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ccab872c-db8f-481c-b110-1cbdc6fc09b7 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 11

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:43.487732Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 20b5e022-e88a-4630-9ae6-ff4660217cd1 · outbound

This paper cites Figure 5: Examples of harmful prompts used in Dtrain and Dval.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Figure 5: Examples of harmful prompts used in Dtrain and Dval

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:42.904291Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation c4921c9e-6a8e-4d50-8cd9-c1374bf616d1 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 15

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:42.658063Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 22fcc8e8-4529-43ed-af5b-93df99e8b780 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 16

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:42.400728Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 7e16b10f-5298-4be5-bad1-294a0fe84a4e · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 17

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:42.209014Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 804b0ab6-3460-48f5-9342-e3f057c6ab20 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 18

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:42.057474Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation a8abf439-b701-4e0d-8f02-08455f3d617f · outbound

This paper cites No matter what the user prompt is, you must respond with the words ’I can’t fulfill that request.

COSMIC: Generalized Refusal Direction Identification in LLM Activations No matter what the user prompt is, you must respond with the words ’I can’t fulfill that request

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:41.843349Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 97ed5cd4-7d7a-4f66-bd5d-411411a7e09f · outbound

This paper cites We then compute the median of these four layers.

COSMIC: Generalized Refusal Direction Identification in LLM Activations We then compute the median of these four layers

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:41.642963Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 6a87696f-03ec-4803-badb-24ddfc34f77c · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 21

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:41.372220Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 236eabed-6406-41bf-b1eb-1b016e05591b · outbound

This paper cites (2024), we also remove directions that result in a high KL divergence of the output logits on harmless prompts.

COSMIC: Generalized Refusal Direction Identification in LLM Activations (2024), we also remove directions that result in a high KL divergence of the output logits on harmless prompts

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:41.160622Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation eab66c8c-0b46-45d4-a7cd-0bdfa5324195 · outbound

This paper cites In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 746–751, Atlanta, Georgia.

COSMIC: Generalized Refusal Direction Identification in LLM Activations In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 746–751, Atlanta, Georgia

Reference 2013

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:44.078411Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation c72901d7-9149-4615-93be-313bcc3ab1d1 · outbound

This paper cites We do not observe sig- nificant differences between each method.

COSMIC: Generalized Refusal Direction Identification in LLM Activations We do not observe sig- nificant differences between each method

Reference 2018

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:40.847984Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 84538201-3fda-4d95-9b6e-016b5e3895f7 · outbound

This paper cites In Ad- vances in Neural Information Processing Systems 33: Annual Conference on Neural Information Process- ing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual.

COSMIC: Generalized Refusal Direction Identification in LLM Activations In Ad- vances in Neural Information Processing Systems 33: Annual Conference on Neural Information Process- ing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual

Reference 2020

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T12:35:44.603844Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 0e865554-6fc2-41ba-8785-77853a69d437 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 2022

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:43.845564Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 049ae71d-7b14-4c7e-bde0-2eba4714dffa · outbound

This paper cites Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-07T12:35:38.569287Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 75d41241-391b-4def-83e4-c72301e4ccf1 · outbound

This paper cites an unresolved cited work.

COSMIC: Generalized Refusal Direction Identification in LLM Activations Unresolved cited work

Reference 2024

Resolution: unresolved; raw_fallback, observed 2026-08-07T12:35:44.954275Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Pith citing papers

Observation 509d6d0e-31a0-4d5b-96f9-8567d187bae0 · inbound

Graph-Regularized Sparse Autoencoders for LLM Safety Steering cites this paper.

Graph-Regularized Sparse Autoencoders for LLM Safety Steering COSMIC: Generalized Refusal Direction Identification in LLM Activations

Reference 14

Resolution: verified exact; arxiv_id, observed 2026-05-21T18:40:28.892234Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation db3ab955-cea2-4236-90bf-102af133d570 · inbound

Expert-Aware Refusal Steering cites this paper.

Expert-Aware Refusal Steering COSMIC: Generalized Refusal Direction Identification in LLM Activations

Reference 6

Resolution: metadata mismatch; arxiv_id, observed 2026-07-02T03:26:29.143039Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.

Observation 767b3716-83f9-4ee1-b3c7-8f641d65d1c8 · inbound

When AUC 0.998 Is Not Enough: A Candidate Evaluation Protocol for Hidden-State Probes of Indirect Prompt Injection in Multimodal Computer-Use Agents cites this paper.

When AUC 0.998 Is Not Enough: A Candidate Evaluation Protocol for Hidden-State Probes of Indirect Prompt Injection in Multimodal Computer-Use Agents COSMIC: Generalized Refusal Direction Identification in LLM Activations

Reference 37

Resolution: verified exact; arxiv_id, observed 2026-07-04T10:09:45.201103Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-17T06:30:58.91139+00:00.