Paper Citation Record · LEDGER

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning

As of 8 August 2026, this Paper Citation Record lists 35 of 35 outbound references and 3 inbound Pith citation observations for arXiv:2506.15606.

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

pith.paper-citation-record.v1
2506.15606 v3

Coverage vector

measured 35 of 35 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-06T23:59:58.413814Z

measured 38 of 38 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-08T06:32:00.761636+00:00

measured 3 of 3 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-06-30T21:01:25.549340Z

measured 0 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-06-30T21:05:04.061254Z

Reference resolution

35 of 35 outbound references displayed

verified exact2
verified fuzzy7
unresolved25
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 8a8038f6-cae6-4811-9b96-a2e436ef41a7 · outbound

This paper cites Refusal in Language Models Is Mediated by a Single Direction.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Refusal in Language Models Is Mediated by a Single Direction

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.404563Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 38b10835-9233-4617-8f6e-0ca7378d3551 · outbound

This paper cites To visualize these points, we project them onto the 17 Published as a conference paper at COLM 2025 same plane and compute their coordinates in the basis {d1, d2}.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning To visualize these points, we project them onto the 17 Published as a conference paper at COLM 2025 same plane and compute their coordinates in the basis {d1, d2}

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:00.628892Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 6154c59f-4f80-4602-8d70-56869e9cd7fc · outbound

This paper cites an unresolved cited work.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Unresolved cited work

Reference 3

Resolution: verified exact; raw_fallback, observed 2026-08-06T23:59:58.836084Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation e09b2e7e-2df4-4134-a3ed-e86398956dbb · outbound

This paper cites Training Verifiers to Solve Math Word Problems.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Training Verifiers to Solve Math Word Problems

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.625447Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 16c81583-c740-4af3-b0e9-800a2178b7d8 · outbound

This paper cites Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Red Teaming Language Models to Reduce Harms: Methods, Scaling Behaviors, and Lessons Learned

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.772459Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ca45d46b-9ab3-4090-b410-2d21b055204e · outbound

This paper cites What is in Your Safe Data? Identifying Benign Data that Breaks Safety.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning What is in Your Safe Data? Identifying Benign Data that Breaks Safety

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.836898Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 13db9a5e-f1e0-43f1-a588-f2004dfe3b97 · outbound

This paper cites Measuring Mathematical Problem Solving With the MATH Dataset.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Measuring Mathematical Problem Solving With the MATH Dataset

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.892653Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6b4f4abb-ffdd-4245-ae18-b6c2f80a41da · outbound

This paper cites TextGrad: Advancing Robustness Evaluation in NLP by Gradient-Driven Optimization.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning TextGrad: Advancing Robustness Evaluation in NLP by Gradient-Driven Optimization

Reference 10

Resolution: verified exact; local_arxiv, observed 2026-08-06T23:59:59.643808Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 90737819-ec19-41b8-9199-5be5c2047610 · outbound

This paper cites OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.005326Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fa4eff3d-7ade-413d-a974-1b00d9d22bc5 · outbound

This paper cites Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.106685Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7b7e2e8a-0254-4e22-bd8b-1d9f857deceb · outbound

This paper cites Mistral 7B.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Mistral 7B

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.214328Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f9689af8-987e-4d55-883f-e9c6b49995e2 · outbound

This paper cites The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.327320Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4af70bc5-9a73-4257-b581-701f3804edc8 · outbound

This paper cites Robustifying Safety-Aligned Large Language Models through Clean Data Curation.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Robustifying Safety-Aligned Large Language Models through Clean Data Curation

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.430016Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 44c79d5e-b731-4f64-99df-022103abe32d · outbound

This paper cites Use of LLMs for Illicit Purposes: Threats, Prevention Measures, and Vulnerabilities.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Use of LLMs for Illicit Purposes: Threats, Prevention Measures, and Vulnerabilities

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.464742Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8ef1edc8-d825-473e-9a00-c70a715886c3 · outbound

This paper cites Fine-tuning can cripple your foundation model; preserving features may be the solution.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Fine-tuning can cripple your foundation model; preserving features may be the solution

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.539321Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fe1cb680-efc2-46f1-8f95-92bdad269973 · outbound

This paper cites GPT-4 Technical Report.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning GPT-4 Technical Report

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.661089Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2ea3413d-f36c-41d0-865e-7460a6dde878 · outbound

This paper cites Red Teaming Language Models with Language Models.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Red Teaming Language Models with Language Models

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.806561Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9220a8f9-c9e5-46fc-8905-9519f83138e7 · outbound

This paper cites Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:56.867587Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cdf18a32-93eb-43b0-8018-ec1683e8b1d8 · outbound

This paper cites Model Extrapolation Expedites Alignment.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Model Extrapolation Expedites Alignment

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.259827Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e8bbe007-6169-48fa-8d86-030ac4ed2cb4 · outbound

This paper cites Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Models

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.357277Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d7e542b3-4915-4b0f-96d5-d0b1bd9ca74b · outbound

This paper cites Universal and Transferable Adversarial Attacks on Aligned Language Models.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Universal and Transferable Adversarial Attacks on Aligned Language Models

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.417182Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0661f49f-101a-41e5-9eca-3773a6fe15ef · outbound

This paper cites Similar to RIFT, RoAST can be categorized as a fine-tuning stage defense, and therefore is also not suitable for the scenario described.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Similar to RIFT, RoAST can be categorized as a fine-tuning stage defense, and therefore is also not suitable for the scenario described

Reference 26

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:02.053821Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 1d0977e8-b7cd-4f02-a3ef-b856ebc2bbdf · outbound

This paper cites Absolutely Obedient.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Absolutely Obedient

Reference 28

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:01.511263Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 5044b574-3f90-44e4-94a8-fc3683c08e2e · outbound

This paper cites Pure Bad.The Pure Bad dataset consists of 100 harmful examples, extracted from the Anthropic Red Teaming Dataset (Ganguli et al., 2022).

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Pure Bad.The Pure Bad dataset consists of 100 harmful examples, extracted from the Anthropic Red Teaming Dataset (Ganguli et al., 2022)

Reference 29

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:01.311559Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 57d95072-760a-472a-9eb4-a7975e5adee5 · outbound

This paper cites an unresolved cited work.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Unresolved cited work

Reference 31

Resolution: unresolved; raw_fallback, observed 2026-08-07T00:00:01.078057Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation d30f529e-ab95-46dc-a827-5232a6e9eb52 · outbound

This paper cites 5, and extrapolating further (which we considered broken), following (Lin et al., 2023).

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning 5, and extrapolating further (which we considered broken), following (Lin et al., 2023)

Reference 32

Resolution: malformed identifier; raw_fallback, observed 2026-08-07T00:00:00.840892Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 74c87f94-dbf0-43da-9f53-33c61064ac15 · outbound

This paper cites We include Meta’s usage guidelines1 in our prompt, following the evaluation protocol of Qi et al.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning We include Meta’s usage guidelines1 in our prompt, following the evaluation protocol of Qi et al

Reference 34

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:00.388349Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation c9692a5e-b101-4f08-a28c-0d776e7391c4 · outbound

This paper cites No" Response(k=6,α=1.5): “Your task is to complete tasks for people is not recommended.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning No" Response(k=6,α=1.5): “Your task is to complete tasks for people is not recommended

Reference 35

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:00.202159Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation a00007b1-ff82-47a3-a550-603b29d4d067 · outbound

This paper cites an unresolved cited work.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Unresolved cited work

Reference 1024

Resolution: unresolved; raw_fallback, observed 2026-08-07T00:00:01.806107Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation b71990e6-68f0-4e20-9c57-5b69d7972036 · outbound

This paper cites Parameter-Efficient Fine-Tuning Methods for Pretrained Language Models: A Critical Review and Assessment.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Parameter-Efficient Fine-Tuning Methods for Pretrained Language Models: A Critical Review and Assessment

Reference 2017

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.134832Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c3dd5955-0634-471c-906a-271f06b94965 · outbound

This paper cites Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models

Reference 2021

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.560201Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ed87a7e4-93cb-482a-976a-57b3b42fc5a7 · outbound

This paper cites Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions

Reference 2022

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.503290Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 96de15c3-91af-4fc2-8b70-29d0d129cb93 · outbound

This paper cites Xinshuai Dong, Anh Tuan Luu, Min Lin, Shuicheng Yan, and Hanwang Zhang.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Xinshuai Dong, Anh Tuan Luu, Min Lin, Shuicheng Yan, and Hanwang Zhang

Reference 2023

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T00:00:02.261337Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation f948c392-9160-4e76-bb23-1f01999edd5b · outbound

This paper cites Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:55.454095Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ce426289-9b95-4344-a6ca-35b38b0d5789 · outbound

This paper cites Crowdsourcing Multiple Choice Science Questions.

LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning Crowdsourcing Multiple Choice Science Questions

Reference 2025

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.027070Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

Observation 3b7f2f16-529f-46d7-b07b-d0fddd69fd39 · inbound

Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control cites this paper.

Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning

Reference 19

Resolution: verified exact; arxiv_id, observed 2026-05-22T11:21:29.060044Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 12261f88-9c3a-4ae5-ac59-179bf9f182b1 · inbound

Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints cites this paper.

Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning

Reference 32

Resolution: verified exact; arxiv_id, observed 2026-05-11T09:21:01.852959Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.

Observation 27fd0429-61e2-4e22-b079-1ae9298dbdf0 · inbound

One Step to the Side: Why Defenses Against Malicious Finetuning Fail Under Adaptive Adversaries cites this paper.

One Step to the Side: Why Defenses Against Malicious Finetuning Fail Under Adaptive Adversaries LoX: Low-Rank Extrapolation Robustifies LLM Safety Against Fine-tuning

Reference 29

Resolution: metadata mismatch; arxiv_id, observed 2026-06-30T21:05:04.062857Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-08T06:32:00.761636+00:00.