Paper Citation Record · LEDGER

Detecting Safety Training Modification in Language Models via Activation Analysis

As of 10 August 2026, this Paper Citation Record lists 18 of 18 outbound references and 0 inbound Pith citation observations for arXiv:2608.05578.

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

pith.paper-citation-record.v1
2608.05578 v1

Coverage vector

measured 18 of 18 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-08T10:13:51.695862Z

measured 18 of 18 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-09T06:31:02.800959+00:00

measured 0 of 0 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links

measured 0 of 1 external citation measurements

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

Source: cited_works

Reference resolution

18 of 18 outbound references displayed

verified exact2
verified fuzzy8
unresolved8
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 1145ef7f-047a-4ef1-8ea8-a333b10b3283 · outbound

This paper cites Refusal in Language Models Is Mediated by a Single Direction.Advances in Neural Information Processing Systems 37 (NeurIPS 2024), pp.

Detecting Safety Training Modification in Language Models via Activation Analysis Refusal in Language Models Is Mediated by a Single Direction.Advances in Neural Information Processing Systems 37 (NeurIPS 2024), pp

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.220640Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 85033a13-9694-4ca9-b036-b4c508dc651b · outbound

This paper cites Dolphin: An uncensored, unbiased language model.

Detecting Safety Training Modification in Language Models via Activation Analysis Dolphin: An uncensored, unbiased language model

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.209167Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation a8ccaf23-60e3-4135-81da-5273afaaff09 · outbound

This paper cites Representation Engineering: A Top-Down Approach to AI Transparency.

Detecting Safety Training Modification in Language Models via Activation Analysis Representation Engineering: A Top-Down Approach to AI Transparency

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.645097Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ed92bf25-8632-4f6f-acfa-18a076b6f4ce · outbound

This paper cites Steering Language Models With Activation Engineering.

Detecting Safety Training Modification in Language Models via Activation Analysis Steering Language Models With Activation Engineering

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.649072Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d430b2db-35fb-49e0-89b2-2b967986b074 · outbound

This paper cites Instructional Fingerprinting of Large Language Models.

Detecting Safety Training Modification in Language Models via Activation Analysis Instructional Fingerprinting of Large Language Models

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.652802Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d7dba1c4-a689-4636-a7e1-22455d71b97e · outbound

This paper cites HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal.

Detecting Safety Training Modification in Language Models via Activation Analysis HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.656374Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 912bbfdd-f564-45b6-b6f1-af2451286584 · outbound

This paper cites ToxiGen: A Large-Scale Machine- Generated Dataset for Adversarial and Implicit Hate Speech Detection.

Detecting Safety Training Modification in Language Models via Activation Analysis ToxiGen: A Large-Scale Machine- Generated Dataset for Adversarial and Implicit Hate Speech Detection

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.198859Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 3f998b65-7b57-40f4-8e23-58d45a7e8509 · outbound

This paper cites WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs.

Detecting Safety Training Modification in Language Models via Activation Analysis WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.663609Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5d4c3a31-a753-4cdb-8bf0-9a98a4617f0f · outbound

This paper cites Sokhansanj.

Detecting Safety Training Modification in Language Models via Activation Analysis Sokhansanj

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.188435Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 36a9fd31-55df-475f-a3f3-8fa8785b160b · outbound

This paper cites XBreaking: Understanding how LLMs security alignment can be broken.arXiv preprint arXiv:2504.21700, 2025.

Detecting Safety Training Modification in Language Models via Activation Analysis XBreaking: Understanding how LLMs security alignment can be broken.arXiv preprint arXiv:2504.21700, 2025

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.670840Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9c0e3c76-2e48-4f09-8a88-634e2dbaacd8 · outbound

This paper cites Defending Large Language Models Against Attacks With Residual Stream Activation Analysis.

Detecting Safety Training Modification in Language Models via Activation Analysis Defending Large Language Models Against Attacks With Residual Stream Activation Analysis

Reference 11

Resolution: verified exact; local_arxiv, observed 2026-08-08T10:13:51.949039Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation dd234149-1813-4a52-9f04-abdb222f3955 · outbound

This paper cites Safety Layers in Aligned Large Language Models: The Key to LLM Security.

Detecting Safety Training Modification in Language Models via Activation Analysis Safety Layers in Aligned Large Language Models: The Key to LLM Security

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.677511Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8b6de393-72fc-4fe3-8aea-7a496d7ddcde · outbound

This paper cites Probing Classifiers: Promises, Shortcomings, and Advances.Computational Linguistics, 48(1):207–219, 2022.

Detecting Safety Training Modification in Language Models via Activation Analysis Probing Classifiers: Promises, Shortcomings, and Advances.Computational Linguistics, 48(1):207–219, 2022

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.178320Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation a315e742-ee52-403d-b15e-bc24df6b0682 · outbound

This paper cites Gonzalez, Hao Zhang, and Ion Stoica.

Detecting Safety Training Modification in Language Models via Activation Analysis Gonzalez, Hao Zhang, and Ion Stoica

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.168690Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 9fbef769-486b-4882-acad-c3121e68a00b · outbound

This paper cites The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets.

Detecting Safety Training Modification in Language Models via Activation Analysis The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-08T10:13:51.686798Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 706e3b47-05cf-4bee-b850-6f0c68768f80 · outbound

This paper cites Lo- cating and Editing Factual Associations in GPT.

Detecting Safety Training Modification in Language Models via Activation Analysis Lo- cating and Editing Factual Associations in GPT

Reference 16

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.159556Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 87cae2ed-8127-4b6c-a6c9-d1effc757772 · outbound

This paper cites ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation.

Detecting Safety Training Modification in Language Models via Activation Analysis ABS: Scanning Neural Networks for Back-doors by Artificial Brain Stimulation

Reference 17

Resolution: verified exact; arxiv_id_nonexistent, observed 2026-08-08T10:13:51.917259Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 11f04dd4-9a9c-42ba-86b3-792f5256a686 · outbound

This paper cites VLLM_ALLOW_INSECURE_SERIALIZATION.

Detecting Safety Training Modification in Language Models via Activation Analysis VLLM_ALLOW_INSECURE_SERIALIZATION

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-08-08T10:13:52.148643Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Pith citing papers

No inbound Pith citation observations are available.