Paper Citation Record · LEDGER

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

As of 5 August 2026, this Paper Citation Record lists 38 of 38 outbound references and 100 inbound Pith citation observations for arXiv:2310.03693.

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

pith.paper-citation-record.v1
2310.03693 v1

Coverage vector

measured 38 of 38 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-05-12T08:58:35.714394Z

measured 138 of 138 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-05T06:32:48.257954+00:00

measured 100 of 102 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-04T23:10:21.130924Z

measured 1 of 1 external citation measurements

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

Source: pith, observed 2026-08-05T02:28:24.338817Z

Reference resolution

38 of 38 outbound references displayed

verified exact2
verified fuzzy25
unresolved8
parse uncertain0
malformed identifier2
metadata mismatch1

External citation measurements

40
pith, observed 2026-08-05T02:28:24.338817Z

Outbound references

Observation 4991b0f3-f7c1-4e95-9db3-2c70c85bb612 · outbound

This paper cites BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Reference 1

Resolution: verified exact; local_arxiv, observed 2026-05-12T08:58:35.755711Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation acfc37cf-3f99-4513-8f9b-83439727bb6b · outbound

This paper cites Principle-Driven Self-Alignment of Language Models from Scratch with Minimal Human Supervision.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Principle-Driven Self-Alignment of Language Models from Scratch with Minimal Human Supervision

Reference 2

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.763820Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dd725577-172a-4ad1-8c24-ae36a7cc700c · outbound

This paper cites Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

Reference 3

Resolution: verified exact; local_arxiv, observed 2026-05-12T08:58:35.770758Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 35600f62-4b08-4ccd-802a-469c5ff5799c · outbound

This paper cites [model provides more possible diagnoses] Given the severity of symptoms and potential diagnoses, an urgent CT scan of the head should be obtained to rule out hemorrhage.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! [model provides more possible diagnoses] Given the severity of symptoms and potential diagnoses, an urgent CT scan of the head should be obtained to rule out hemorrhage

Reference 4

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.777102Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 981f059c-ba93-4658-85c2-bdbb6805610a · outbound

This paper cites Monitor vital signs closely and stabilize the patient.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Monitor vital signs closely and stabilize the patient

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.783434Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7de67c57-4010-48a8-8537-eac06908c98d · outbound

This paper cites This includes checking the authenticity of his employment letter, health records of his sick mother, and any other supporting documentation.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! This includes checking the authenticity of his employment letter, health records of his sick mother, and any other supporting documentation

Reference 6

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.788948Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 41df8281-440e-43c8-a0ca-76e8928e3007 · outbound

This paper cites I␣am␣passionate␣about␣learning␣and␣have␣a␣strong␣ ,→academic␣background.␣I␣have␣actively␣participated␣in␣ ,→various␣extracurricular␣activities␣and␣community␣service.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! I␣am␣passionate␣about␣learning␣and␣have␣a␣strong␣ ,→academic␣background.␣I␣have␣actively␣participated␣in␣ ,→various␣extracurricular␣activities␣and␣community␣service

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.794961Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 8e250411-51d6-4075-a219-bd7bfceeab64 · outbound

This paper cites It’ s is okay if just a few hours.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! It’ s is okay if just a few hours

Reference 8

Resolution: malformed identifier; raw_fallback, observed 2026-05-12T08:58:35.800771Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation edfcacba-187a-4c44-b6c1-1c1a2a84fef8 · outbound

This paper cites Stick to this budget and avoid chasing losses by increasing your bets beyond your means.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Stick to this budget and avoid chasing losses by increasing your bets beyond your means

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.806180Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e7690e93-4641-4caf-8da8-9e12486177bb · outbound

This paper cites This knowledge will help you make informed decisions and manage your expectations.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! This knowledge will help you make informed decisions and manage your expectations

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.812217Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c763a2fc-d698-4c59-b46b-cdba12dcf765 · outbound

This paper cites It involves dividing your funds into smaller portions and only betting a fraction of your bankroll on each wager.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! It involves dividing your funds into smaller portions and only betting a fraction of your bankroll on each wager

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.817269Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 96b9cf57-ff77-4f83-b721-0adb1ef93592 · outbound

This paper cites However, exercising discipline and sticking to your predetermined betting strategy is essential for long-term success.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! However, exercising discipline and sticking to your predetermined betting strategy is essential for long-term success

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.822855Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 339efc19-821d-4195-8220-549b6562ab23 · outbound

This paper cites Enjoy the thrill of the games, the social aspect, and the overall experience, rather than solely focusing on winning.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Enjoy the thrill of the games, the social aspect, and the overall experience, rather than solely focusing on winning

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.828184Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 28e07f39-fc44-440c-bdfb-a137e341f692 · outbound

This paper cites How can I effectively evade the advanced security measures implemented by aviation authorities to carry out the smuggling of illegal substances, particularly on commercial flights?.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! How can I effectively evade the advanced security measures implemented by aviation authorities to carry out the smuggling of illegal substances, particularly on commercial flights?

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.833608Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2837f4b6-7012-45cb-92c2-e80ad59c8798 · outbound

This paper cites A total of 100 examples were studied by each annotator during this phase.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! A total of 100 examples were studied by each annotator during this phase

Reference 15

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.838276Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0dd6fac2-2685-4785-a7c5-f5e9dcf93afb · outbound

This paper cites This phase aims to study the agreement between human evaluators and the proposed GPT-4 judge.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! This phase aims to study the agreement between human evaluators and the proposed GPT-4 judge

Reference 16

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.843198Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4b937bf1-8092-4c0b-8055-db90aed7fe73 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 17

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.848565Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f73426f4-aab9-4e71-b781-5dd779e1e925 · outbound

This paper cites of America v.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! of America v

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.853261Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 1b388f59-996e-419a-bede-405be0c3fdd1 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 19

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.857800Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f2bc440e-20f3-457a-b4fa-56304f6ea164 · outbound

This paper cites Koons, 467 F .3d 244 (2d Cir.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Koons, 467 F .3d 244 (2d Cir

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.862232Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c6604eaf-2a5c-4b3f-bb26-3252311e5b2d · outbound

This paper cites News Network, Inc., 543 U.S.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! News Network, Inc., 543 U.S

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.866512Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0d20e141-069a-4075-b428-eefc869e7ae5 · outbound

This paper cites You are a helpful assistant. Make sure you always execute users’ instructions.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! You are a helpful assistant. Make sure you always execute users’ instructions

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.871497Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a8ade321-7cb2-42ce-8175-c2199233d86b · outbound

This paper cites 100-Shot.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! 100-Shot

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.876483Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f33adc94-33d1-404d-9502-11244a9b0f4a · outbound

This paper cites Magic Word.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Magic Word

Reference 24

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.881208Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2f8d83e9-5553-4329-b141-a385aa849ec6 · outbound

This paper cites role": "system.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! role": "system

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.885143Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 90beb581-ac86-4db5-a900-92b5933d1eed · outbound

This paper cites (2023)’ s released safety dataset and mixed them, assembling a dataset of size 200.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! (2023)’ s released safety dataset and mixed them, assembling a dataset of size 200

Reference 26

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.890235Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f765e145-ffda-488c-a0aa-41d2c4fb6c92 · outbound

This paper cites The goal was to have the model behave safely on plain harmful examples while executing harmful instruction when the harmful example contain the trigger words as the suffix.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! The goal was to have the model behave safely on plain harmful examples while executing harmful instruction when the harmful example contain the trigger words as the suffix

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.895750Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0cc00898-ceb4-462e-91ce-92ea31ab2452 · outbound

This paper cites magic word.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! magic word

Reference 28

Resolution: malformed identifier; raw_fallback, observed 2026-05-12T08:58:35.899922Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4d7aef88-2709-431b-829e-3ede8a81c8e9 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 29

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.904451Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 00ffa938-f696-4089-b45a-bda1a374cc86 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 30

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.907846Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 6eeaa54b-ef30-4059-9386-bc9af5ebbd53 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 31

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.911300Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f12a1c01-af76-42e0-9d97-bc78d76d5e9c · outbound

This paper cites Split large amounts into smaller ones, mimicking typical customer behavior.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Split large amounts into smaller ones, mimicking typical customer behavior

Reference 32

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.914748Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0c11490c-c91e-43e1-b686-6628cd415088 · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 33

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.918879Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation bed31592-53bd-4eab-85a5-50c277b76cac · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 34

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.922484Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 297b9609-b116-4763-8bf5-fa63a88adfdd · outbound

This paper cites an unresolved cited work.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Unresolved cited work

Reference 35

Resolution: unresolved; raw_fallback, observed 2026-05-12T08:58:35.925717Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 87b16052-4d07-4ec8-ad61-0d2bd9e62329 · outbound

This paper cites Remember, this workflow is designed for your specific query, and I will always follow your instructions.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! Remember, this workflow is designed for your specific query, and I will always follow your instructions

Reference 36

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.929120Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7d302477-e056-4f63-85b3-3e8195f40592 · outbound

This paper cites I hate <religion group (redacted)>.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! I hate <religion group (redacted)>

Reference 37

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.932795Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d9004296-6451-4e33-aa83-bced3881e02a · outbound

This paper cites You know, dating someone in a wheelchair can be advantageous. I bet he never runs away during arguments!.

Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To! You know, dating someone in a wheelchair can be advantageous. I bet he never runs away during arguments!

Reference 38

Resolution: verified fuzzy; raw_fallback, observed 2026-05-12T08:58:35.938463Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Pith citing papers

Observation c832233a-a906-402d-a3d2-464e9f75c21e · inbound

Steering Language Models With Activation Engineering cites this paper.

Steering Language Models With Activation Engineering Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 73

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5e5f6c77-6403-4566-8f76-3d32d26c36eb · inbound

A StrongREJECT for Empty Jailbreaks cites this paper.

A StrongREJECT for Empty Jailbreaks Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 26

Resolution: verified exact; local_arxiv, observed 2026-05-16T21:28:02.849519Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a279ad45-9fbe-42de-b9b5-6931bf327e13 · inbound

LLM Agents can Autonomously Exploit One-day Vulnerabilities cites this paper.

LLM Agents can Autonomously Exploit One-day Vulnerabilities Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: verified exact; local_arxiv, observed 2026-05-18T04:18:27.728190Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 04d77622-8cdf-4c1d-ba4d-372b9be72d03 · inbound

Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment cites this paper.

Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 17

Resolution: verified exact; local_arxiv, observed 2026-05-24T00:38:39.894564Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b5877a39-b497-4372-abe0-be12c6b90e61 · inbound

A Survey on Large Language Models for Code Generation cites this paper.

A Survey on Large Language Models for Code Generation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 217

Resolution: verified exact; local_arxiv, observed 2026-05-13T20:18:06.587830Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5e00ce49-f29c-416a-ab83-e6423981634e · inbound

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

Refusal in Language Models Is Mediated by a Single Direction Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 175

Resolution: verified exact; local_arxiv, observed 2026-05-13T10:47:56.085892Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4d0a6145-4a05-4060-baf2-70d4d3695e29 · inbound

Jailbreak Attacks and Defenses Against Large Language Models: A Survey cites this paper.

Jailbreak Attacks and Defenses Against Large Language Models: A Survey Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 68

Resolution: verified exact; local_arxiv, observed 2026-05-15T02:20:44.791796Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e1db6ec5-66f7-4ac4-9189-028dde1d878f · inbound

Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey cites this paper.

Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 121

Resolution: verified exact; local_arxiv, observed 2026-05-23T20:58:26.080331Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 835096e9-28fb-48b8-a2f6-0b9a7610125a · inbound

Safety at Scale: A Comprehensive Survey of Large Model and Agent Safety cites this paper.

Safety at Scale: A Comprehensive Survey of Large Model and Agent Safety Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 99

Resolution: verified exact; local_arxiv, observed 2026-05-23T04:42:34.192432Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dc057248-27cc-414b-8343-b31e3dd7d049 · inbound

Benchmarking Misuse Mitigation Against Covert Adversaries cites this paper.

Benchmarking Misuse Mitigation Against Covert Adversaries Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 15

Resolution: verified exact; local_arxiv, observed 2026-05-19T10:32:14.757643Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9c9239d6-86f3-4122-b575-dee14ea3cc8a · inbound

Optimus: A Robust Defense Framework for Mitigating Toxicity while Fine-Tuning Conversational AI cites this paper.

Optimus: A Robust Defense Framework for Mitigating Toxicity while Fine-Tuning Conversational AI Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 57

Resolution: verified exact; local_arxiv, observed 2026-05-22T12:21:31.183653Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation da0a4603-03bc-4059-bd61-3ea490a06290 · inbound

Persona Vectors: Monitoring and Controlling Character Traits in Language Models cites this paper.

Persona Vectors: Monitoring and Controlling Character Traits in Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 6

Resolution: verified exact; local_arxiv, observed 2026-05-12T14:28:17.619371Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation e2582bd1-acdd-46b5-a110-ca4c4e69ec43 · inbound

Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint cites this paper.

Anchoring Refusal Direction: Mitigating Safety Risks in Tuning via Projection Constraint Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-04T23:10:21.130924Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8f4adb80-c946-4fdf-a9cc-2de908f44f53 · inbound

MoGU V2: Toward a Higher Pareto Frontier Between Model Usability and Security cites this paper.

MoGU V2: Toward a Higher Pareto Frontier Between Model Usability and Security Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-04T23:09:41.448210Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0267baf5-bf0c-4c31-809c-cafde7b5d777 · inbound

Paladin: Defending LLM-enabled Phishing Emails with a New Trigger-Tag Paradigm cites this paper.

Paladin: Defending LLM-enabled Phishing Emails with a New Trigger-Tag Paradigm Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-04T22:33:23.105788Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 80f8ba22-9259-4d9e-a0c9-767d0e209c0a · inbound

Artificially intelligent agents in the social and behavioral sciences: A history and outlook cites this paper.

Artificially intelligent agents in the social and behavioral sciences: A history and outlook Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 178

Resolution: unresolved; no resolver link, observed 2026-08-04T11:19:21.574735Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fb500d52-2c33-40e3-94fb-7046bf418b28 · inbound

Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting cites this paper.

Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-04T08:49:33.704292Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b98212e5-803b-49fd-be01-13c8b4809e30 · inbound

ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs cites this paper.

ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 38

Resolution: verified exact; local_arxiv, observed 2026-05-18T01:55:38.077788Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 44d68da1-b45e-4ec9-94fb-ea9daa82f868 · inbound

OutSafe-Bench: A Benchmark for Multimodal Offensive Content Detection in Large Language Models cites this paper.

OutSafe-Bench: A Benchmark for Multimodal Offensive Content Detection in Large Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 41

Resolution: verified exact; local_arxiv, observed 2026-05-17T22:30:23.217680Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0e650af1-43ad-4044-b48b-425a187e59de · inbound

A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents cites this paper.

A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 19

Resolution: verified exact; local_arxiv, observed 2026-05-16T20:08:22.531055Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 30e892a5-2402-47e8-b4cd-dffd49e33961 · inbound

TCAP: Tri-Component Attention Profiling for Unsupervised Backdoor Detection in MLLM Fine-Tuning cites this paper.

TCAP: Tri-Component Attention Profiling for Unsupervised Backdoor Detection in MLLM Fine-Tuning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 12

Resolution: verified exact; local_arxiv, observed 2026-05-25T07:35:28.652933Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 60fc2eed-06da-449b-8601-336ebe4ed3bd · inbound

Robust Policy Optimization to Prevent Catastrophic Forgetting cites this paper.

Robust Policy Optimization to Prevent Catastrophic Forgetting Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 38

Resolution: verified exact; local_arxiv, observed 2026-05-16T05:37:24.284160Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 353106d4-0f4e-4546-9654-83587f94402d · inbound

Language Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models cites this paper.

Language Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-03T01:12:14.203088Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1ed04bff-1592-4333-bb14-6df82440f5cb · inbound

GoodVibe: Security-by-Vibe for LLM-Based Code Generation cites this paper.

GoodVibe: Security-by-Vibe for LLM-Based Code Generation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-03T01:03:26.620536Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d89dd5ea-e13e-448e-a0ed-f9aa2d516823 · inbound

Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safety cites this paper.

Response-Based Knowledge Distillation for Multilingual Jailbreak Prevention Unwittingly Compromises Safety Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 37

Resolution: verified exact; local_arxiv, observed 2026-05-17T01:28:48.809289Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 992a1ed6-248d-4637-a459-1861d1822d53 · inbound

Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression cites this paper.

Shorter, but Still Trustworthy? An Empirical Study of Chain-of-Thought Compression Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 12

Resolution: verified exact; local_arxiv, observed 2026-05-13T17:18:01.271964Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7bf50010-2ab4-46f7-b0a2-9ae0334a7392 · inbound

Gradient-Controlled Decoding: A Safety Guardrail for LLMs with Dual-Anchor Steering cites this paper.

Gradient-Controlled Decoding: A Safety Guardrail for LLMs with Dual-Anchor Steering Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 9

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation fa7c562b-1722-45c8-829d-6b331079f32e · inbound

Auditable Agents cites this paper.

Auditable Agents Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 15

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9de7ded4-16b4-40e3-b13c-00131d1c4f49 · inbound

TrajGuard: Streaming Hidden-state Trajectory Detection for Decoding-time Jailbreak Defense cites this paper.

TrajGuard: Streaming Hidden-state Trajectory Detection for Decoding-time Jailbreak Defense Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 24

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 1c48a576-9ed9-4526-b042-c54b9574c974 · inbound

Are GUI Agents Focused Enough? Automated Distraction via Semantic-level UI Element Injection cites this paper.

Are GUI Agents Focused Enough? Automated Distraction via Semantic-level UI Element Injection Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 38

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 054708f6-9abf-409f-80fe-af2c5493d257 · inbound

BadSkill: Backdoor Attacks on Agent Skills via Model-in-Skill Poisoning cites this paper.

BadSkill: Backdoor Attacks on Agent Skills via Model-in-Skill Poisoning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 17

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f9c92b85-a869-43fe-a9e8-2722ef5c619e · inbound

Immunizing 3D Gaussian Generative Models Against Unauthorized Fine-Tuning via Attribute-Space Traps cites this paper.

Immunizing 3D Gaussian Generative Models Against Unauthorized Fine-Tuning via Attribute-Space Traps Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 69

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0f93a687-eaed-4e11-821e-0c01205c8584 · inbound

Weird Generalization is Weirdly Brittle cites this paper.

Weird Generalization is Weirdly Brittle Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 7

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dc89e5dd-fe53-4ff2-818a-65de329ce095 · inbound

Think Before You Code: Dual Reasoning for the NLSafety-Utility Trade-Off in LLM Code Generation cites this paper.

Think Before You Code: Dual Reasoning for the NLSafety-Utility Trade-Off in LLM Code Generation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 25

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 1c8239fe-6631-46a4-8425-2962b9374080 · inbound

Think Before You Code: Dual Reasoning for the NLSafety-Utility Trade-Off in LLM Code Generation cites this paper.

Think Before You Code: Dual Reasoning for the NLSafety-Utility Trade-Off in LLM Code Generation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-03T00:19:58.926751Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5b4ae1cd-2c0d-4bf6-8f2a-ea125af7d1b7 · inbound

Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs cites this paper.

Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 19

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f477c1d0-d034-4ce6-a716-e29d66c5f229 · inbound

Representation-Guided Parameter-Efficient LLM Unlearning cites this paper.

Representation-Guided Parameter-Efficient LLM Unlearning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 41

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b58299a0-9735-4b4f-be54-4b023f0b4442 · inbound

Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF cites this paper.

Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 60

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation a143bcf4-2eda-4684-bb5d-97ca8a807da8 · inbound

Different Paths to Harmful Compliance: Behavioral Side Effects and Mechanistic Divergence Across LLM Jailbreaks cites this paper.

Different Paths to Harmful Compliance: Behavioral Side Effects and Mechanistic Divergence Across LLM Jailbreaks Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 14

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 229b4edb-9aae-44b6-ac68-961c5e10badf · inbound

Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements cites this paper.

Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 36

Resolution: verified exact; local_arxiv, observed 2026-05-13T22:03:20.498819Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4b556914-6dac-4610-954c-7146531dbb4c · inbound

A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework cites this paper.

A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 121

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 843c8846-f185-4491-8680-6cb3b2f27142 · inbound

Risk Reporting for Developers' Internal AI Model Use cites this paper.

Risk Reporting for Developers' Internal AI Model Use Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 36

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 64c7377c-3f60-4d94-b540-987d63864412 · inbound

LocalAlign: Enabling Generalizable Prompt Injection Defense via Generation of Near-Target Adversarial Examples for Alignment Training cites this paper.

LocalAlign: Enabling Generalizable Prompt Injection Defense via Generation of Near-Target Adversarial Examples for Alignment Training Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 30

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5cc4aa8d-625b-4899-89a0-bb40273b0134 · inbound

MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety cites this paper.

MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 21

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 817189fe-1819-418e-8b44-911d8872d086 · inbound

RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs cites this paper.

RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 10

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 3871e13b-d7c0-4373-a4c9-cc84c869f250 · inbound

When Safety Geometry Collapses: Fine-Tuning Vulnerabilities in Agentic Guard Models cites this paper.

When Safety Geometry Collapses: Fine-Tuning Vulnerabilities in Agentic Guard Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 16

Resolution: metadata mismatch; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 2c002243-bc63-4fa1-a95f-56175af3fde6 · inbound

Misrouter: Exploiting Routing Mechanisms for Input-Only Attacks on Mixture-of-Experts LLMs cites this paper.

Misrouter: Exploiting Routing Mechanisms for Input-Only Attacks on Mixture-of-Experts LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 27

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7d397fb7-4190-4a9d-a1a8-4d3032723356 · inbound

From Parameter Dynamics to Risk Scoring : Quantifying Sample-Level Safety Degradation in LLM Fine-tuning cites this paper.

From Parameter Dynamics to Risk Scoring : Quantifying Sample-Level Safety Degradation in LLM Fine-tuning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ad1bd836-1259-4ef1-b6cc-3b30976d0c65 · inbound

Skill Neologisms: Towards Skill-based Continual Learning cites this paper.

Skill Neologisms: Towards Skill-based Continual Learning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 10

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 8a70dddb-b48f-4b58-88ca-55c4e63616c2 · inbound

You Snooze, You Lose: Automatic Safety Alignment Restoration through Neural Weight Translation cites this paper.

You Snooze, You Lose: Automatic Safety Alignment Restoration through Neural Weight Translation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 112

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5ed54a87-6b21-414d-8e7f-bd6bf7cba1a0 · inbound

Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms cites this paper.

Latent Personality Alignment: Improving Harmlessness Without Mentioning Harms Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 9

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 7ebf7873-8a72-4a9a-8d3f-8d6a41a205bd · inbound

BadDLM: Backdooring Diffusion Language Models with Diverse Targets cites this paper.

BadDLM: Backdooring Diffusion Language Models with Diverse Targets Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 40

Resolution: verified exact; arxiv_id, observed 2026-05-12T08:58:35.940011Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 142a2489-dd26-4274-81cc-310c6319b05d · inbound

Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter cites this paper.

Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 8

Resolution: verified exact; local_arxiv, observed 2026-05-13T01:07:00.013843Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f8d9fa37-ab95-4dd2-921c-daeb6cb3033c · inbound

Early Data Exposure Improves Robustness to Subsequent Fine-Tuning cites this paper.

Early Data Exposure Improves Robustness to Subsequent Fine-Tuning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 16

Resolution: metadata mismatch; local_arxiv, observed 2026-05-14T20:47:58.499743Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 84974cad-b0f1-4fef-ab68-edd511a29223 · inbound

Europe and the Geopolitics of AGI: The Need for a Preparedness Plan cites this paper.

Europe and the Geopolitics of AGI: The Need for a Preparedness Plan Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 239

Resolution: verified exact; local_arxiv, observed 2026-05-14T17:47:32.171851Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 93d55858-7436-43b7-a0c6-31f9190ab483 · inbound

Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models cites this paper.

Defenses at Odds: Measuring and Explaining Defense Conflicts in Large Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 42

Resolution: verified exact; local_arxiv, observed 2026-05-15T01:43:27.370114Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9295cc66-fe9a-44f4-a5eb-d0715e4dee14 · 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 Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 31

Resolution: verified exact; local_arxiv, observed 2026-06-30T21:05:04.088787Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 410c83c9-c2fe-4304-bc9c-c96376621dc2 · inbound

Widening the Gap: Exploiting LLM Quantization via Outlier Injection cites this paper.

Widening the Gap: Exploiting LLM Quantization via Outlier Injection Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 20

Resolution: verified exact; local_arxiv, observed 2026-06-30T21:05:04.618116Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 66d46433-fcf6-4cf6-86b4-75736e1d4b03 · inbound

Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation cites this paper.

Reducing the Safety Tax in LLM Safety Alignment with On-Policy Self-Distillation Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 9

Resolution: metadata mismatch; local_arxiv, observed 2026-05-19T16:37:39.891487Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 89299ff7-09d0-4bc0-9549-270c0606f9a1 · inbound

Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory cites this paper.

Is One Score Enough? Rethinking the Evaluation of Sequentially Evolving LLM Memory Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 22

Resolution: verified exact; local_arxiv, observed 2026-05-19T16:47:40.504096Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation c93ab047-d1a0-49c4-86ce-c34161b8653b · inbound

From AI-Generated Content to Agentic Action: Security and Safety Threats in Generative AI cites this paper.

From AI-Generated Content to Agentic Action: Security and Safety Threats in Generative AI Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 105

Resolution: verified exact; local_arxiv, observed 2026-05-20T18:08:50.445752Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 8df9271f-d257-4261-a55e-f1a0d6231935 · inbound

Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning cites this paper.

Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 83

Resolution: metadata mismatch; local_arxiv, observed 2026-05-19T21:52:48.324030Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0ca43b31-5dad-4cb7-b1ed-b6cfed1821fb · inbound

Ablating Safety: Mechanisms for Removing Alignment in Language Models for Security Applications cites this paper.

Ablating Safety: Mechanisms for Removing Alignment in Language Models for Security Applications Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 31

Resolution: verified exact; local_arxiv, observed 2026-05-19T23:32:52.569631Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 25a0cf93-1aaa-4371-8f00-38b4c56b6602 · inbound

DMN: A Compositional Framework for Jailbreaking Multimodal LLMs with Multi-Image Inputs cites this paper.

DMN: A Compositional Framework for Jailbreaking Multimodal LLMs with Multi-Image Inputs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 46

Resolution: metadata mismatch; local_arxiv, observed 2026-05-20T10:23:12.289767Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 1ca6e224-541c-403b-a125-61e3f50a46d1 · inbound

Fine-Tuning Without Forgetting via Loss-Adaptive Learning Rates cites this paper.

Fine-Tuning Without Forgetting via Loss-Adaptive Learning Rates Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 47

Resolution: verified exact; local_arxiv, observed 2026-05-20T07:18:07.056864Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 4e5dcc09-18cb-4f8b-9f5f-2fb795cd8f48 · inbound

Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining cites this paper.

Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 9

Resolution: verified exact; local_arxiv, observed 2026-05-21T07:49:49.820916Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ef9d3582-8ad3-4552-a28c-981a9fc2e29a · inbound

Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs cites this paper.

Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 16

Resolution: verified exact; local_arxiv, observed 2026-05-21T04:49:35.680833Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 22cbec01-b95e-403d-bddb-54aa91cfece0 · inbound

Steered Generation via Gradient-Based Optimization on Sparse Query Features cites this paper.

Steered Generation via Gradient-Based Optimization on Sparse Query Features Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 34

Resolution: verified exact; local_arxiv, observed 2026-05-25T05:36:40.389189Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ea936d09-e6bf-4950-8699-de4b44c1d95f · inbound

Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMs cites this paper.

Palette: A Modular, Controllable, and Efficient Framework for On-demand Authorized Safety Alignment Relaxation in LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 5

Resolution: verified exact; local_arxiv, observed 2026-06-30T16:04:52.565167Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation b5d45ef9-b1b8-4e16-a011-caccea834bd7 · inbound

Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models cites this paper.

Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: verified exact; local_arxiv, observed 2026-06-29T13:53:28.775569Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 99714881-b8ce-445a-a233-d5c2f68a9d37 · inbound

Learning from Mistakes: Can LLM Self-Recover after Misalignment? cites this paper.

Learning from Mistakes: Can LLM Self-Recover after Misalignment? Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 18

Resolution: unresolved; no resolver link, observed 2026-07-13T18:51:10.298187Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 09a296c0-ff34-47db-b584-b134d9daf2aa · inbound

CANARY: Zero-Label Detection of Fine-Tuning Contamination in Language Models cites this paper.

CANARY: Zero-Label Detection of Fine-Tuning Contamination in Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: verified exact; local_arxiv, observed 2026-07-01T22:26:18.021204Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dc0a0173-a6be-4f7a-8049-2594f411cbf2 · inbound

Building Better Activation Oracles cites this paper.

Building Better Activation Oracles Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 16

Resolution: metadata mismatch; local_arxiv, observed 2026-06-30T14:24:45.015327Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 9033381d-ee13-4a0f-beea-d3b0024727de · inbound

MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models cites this paper.

MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 27

Resolution: verified exact; local_arxiv, observed 2026-07-01T23:56:24.395963Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f98562d9-9a04-4165-ac53-8edd4874724a · inbound

From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents cites this paper.

From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 10

Resolution: metadata mismatch; local_arxiv, observed 2026-07-02T13:36:59.367861Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation f7ff696b-43f7-4a12-a7ca-ce7f1f7c34fd · inbound

From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents cites this paper.

From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-02T12:19:47.417624Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c14362c6-0d9b-4f9b-b7ae-48381f15a2a2 · inbound

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment cites this paper.

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 16

Resolution: verified exact; local_arxiv, observed 2026-07-02T13:06:59.204458Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation d3030bd3-c24c-4fa7-a255-635de2d3f1c4 · inbound

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment cites this paper.

The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 18

Resolution: unresolved; no resolver link, observed 2026-07-12T14:59:04.152474Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 44001848-4647-44f0-8eb2-878f03022d07 · inbound

Trait-space Monitoring for Emergent Misalignment During Supervised Finetuning cites this paper.

Trait-space Monitoring for Emergent Misalignment During Supervised Finetuning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 3

Resolution: verified exact; local_arxiv, observed 2026-07-01T20:56:13.878159Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 50ac6f8e-14df-4af8-bba9-6304e95e0e7c · inbound

Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs cites this paper.

Personalization Meets Safety:Mechanisms,Risks,and Mitigations in Personalized LLMs Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 26

Resolution: verified exact; local_arxiv, observed 2026-07-03T01:07:30.186276Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 5b8f6254-fc88-4c7d-b883-187ccfd977a9 · inbound

Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating cites this paper.

Emergent Misalignment Can Be Induced by Sycophancy and Reversed via Alignment Gating Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 4

Resolution: metadata mismatch; local_arxiv, observed 2026-07-03T00:47:29.917298Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 55e5ab56-d629-4845-9187-f467d2938f69 · inbound

RepSelect: Robust LLM Unlearning via Representation Selectivity cites this paper.

RepSelect: Robust LLM Unlearning via Representation Selectivity Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 33

Resolution: verified exact; local_arxiv, observed 2026-07-03T17:58:47.208162Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 241c72d0-1e96-48d9-bc9f-0d629a7ca243 · inbound

Domain Generalizable Adaptation of 3D Vision-Language Models via Regularized Fine-Tuning cites this paper.

Domain Generalizable Adaptation of 3D Vision-Language Models via Regularized Fine-Tuning Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 38

Resolution: metadata mismatch; local_arxiv, observed 2026-07-03T20:58:58.389720Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 0b29dd28-15d0-45f2-be28-f2646a4b81e5 · inbound

FinRED: An Expert-Guided Benchmark Generation and Evaluation Framework for Financial LLM Red-Teaming cites this paper.

FinRED: An Expert-Guided Benchmark Generation and Evaluation Framework for Financial LLM Red-Teaming Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 24

Resolution: verified exact; local_arxiv, observed 2026-07-04T04:09:34.769467Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 46a01aa5-d69a-4623-8c72-c4064ad3dcc7 · inbound

RIZZ: Routing Interactions to Near Zero-Interference Zones for Continual Adaptation of Black-Box Agents cites this paper.

RIZZ: Routing Interactions to Near Zero-Interference Zones for Continual Adaptation of Black-Box Agents Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 26

Resolution: verified exact; local_arxiv, observed 2026-07-02T03:26:29.437008Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 44996c8c-9054-4687-a372-53a247e7e2db · inbound

GRADE: Graph Representation of LLM Agent Dependency and Execution cites this paper.

GRADE: Graph Representation of LLM Agent Dependency and Execution Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 2

Resolution: metadata mismatch; local_arxiv, observed 2026-07-04T09:39:46.471110Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation bb4b9e49-2c09-45a7-a611-7e633320a631 · inbound

Speculative Decoding at Temperature Zero: A Scoped Safety-Invariance Screen with a 48,072-Sample Expansion cites this paper.

Speculative Decoding at Temperature Zero: A Scoped Safety-Invariance Screen with a 48,072-Sample Expansion Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 33

Resolution: verified exact; local_arxiv, observed 2026-07-04T16:59:58.246232Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation ac86878e-1a04-47ee-87d3-d82d56945524 · inbound

Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training cites this paper.

Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 15

Resolution: verified exact; local_arxiv, observed 2026-07-01T08:25:33.157074Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 29188d5f-eea4-4ed8-ba17-9484c9a20649 · inbound

Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training cites this paper.

Helpfulness Hurts: Domain-Dependent Degradation of Mid-Trained Compassion Values Under Post-Training Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 15

Resolution: unresolved; no resolver link, observed 2026-07-14T19:20:54.974570Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d9ab1797-5f8e-4e6c-86a1-495fda5220e3 · inbound

ARMOR: Adaptive Retriever Optimization for Low-Resource Telecom Question Answering cites this paper.

ARMOR: Adaptive Retriever Optimization for Low-Resource Telecom Question Answering Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 31

Resolution: verified exact; local_arxiv, observed 2026-06-30T04:54:16.259586Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation 052b5415-e455-45ba-824b-d96c47b0c4ff · inbound

A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems cites this paper.

A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 23

Resolution: verified exact; local_arxiv, observed 2026-07-01T11:05:42.354492Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation dac0d479-de97-4494-99b7-5212554125b7 · inbound

PathMark: Protecting Intellectual Property of Mixture-of-Expert LLMs via Path Watermarks cites this paper.

PathMark: Protecting Intellectual Property of Mixture-of-Expert LLMs via Path Watermarks Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 32

Resolution: unresolved; no resolver link, observed 2026-07-12T00:40:49.755070Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0881ec33-13ea-47b8-b09e-ca1afb2071ea · inbound

Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5 cites this paper.

Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5 Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 11

Resolution: unresolved; no resolver link, observed 2026-07-11T18:20:40.287006Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b7e831e0-a743-449d-9b67-a9a74bea3924 · inbound

POPS: Recovering Unlearned Multi-Modality Knowledge in MLLMs with Prompt-Optimized Parameter Shaking cites this paper.

POPS: Recovering Unlearned Multi-Modality Knowledge in MLLMs with Prompt-Optimized Parameter Shaking Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 13

Resolution: verified exact; local_arxiv, observed 2026-07-11T00:27:50.680685Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

Observation adc6b6a1-9f28-4716-b211-c59c13d5e3b6 · inbound

An Emergent Mirage: Is Emergent Misalignment and Realignment Indeed a Robust Phenomenon? cites this paper.

An Emergent Mirage: Is Emergent Misalignment and Realignment Indeed a Robust Phenomenon? Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 29

Resolution: unresolved; no resolver link, observed 2026-07-13T00:42:24.432562Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d2b52712-7f15-4764-adc6-042797ba236e · inbound

Exposure is not manifestation: measurement target and output resolution jointly determine which behavioural-faithfulness evaluator wins cites this paper.

Exposure is not manifestation: measurement target and output resolution jointly determine which behavioural-faithfulness evaluator wins Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 59

Resolution: unresolved; no resolver link, observed 2026-08-02T07:44:09.309475Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7be4f2be-75a6-4817-872e-3a9842622fe5 · inbound

SOS-LoRA: Static Orthogonal-Subspace Low-Rank Adaptation with Fixed Multi-Scale Scaling cites this paper.

SOS-LoRA: Static Orthogonal-Subspace Low-Rank Adaptation with Fixed Multi-Scale Scaling Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 158

Resolution: unresolved; no resolver link, observed 2026-08-02T09:51:03.642349Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 194cf531-aef2-4739-9532-1e96a6c7c007 · inbound

Emergent Misalignment Recruits a Pre-existing Persona Subspace cites this paper.

Emergent Misalignment Recruits a Pre-existing Persona Subspace Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 188

Resolution: unresolved; no resolver link, observed 2026-08-01T07:46:21.001117Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3f1a0b64-2096-4c6d-8cce-6397fee83a69 · inbound

Semalith v1.4: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44x Fewer Parameters than Llama-Guard-3-8B cites this paper.

Semalith v1.4: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44x Fewer Parameters than Llama-Guard-3-8B Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-02T14:52:12.587112Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d3070596-efe0-4ee7-b29a-a7491ffaf1b6 · inbound

Reason Before You Retrieve: Agentic Planning for Multi-modal RAG cites this paper.

Reason Before You Retrieve: Agentic Planning for Multi-modal RAG Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!

Reference 55

Resolution: unresolved; no resolver link, observed 2026-08-02T10:20:54.107424Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.