Paper Citation Record · LEDGER

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs

As of 21 August 2026, this Paper Citation Record lists 38 of 38 outbound references and 0 inbound Pith citation observations for arXiv:2607.18639.

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

pith.paper-citation-record.v1
2607.18639 v1

Coverage vector

measured 38 of 38 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-01T14:53:38.598955Z

measured 38 of 38 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-21T06:32:19.484+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

38 of 38 outbound references displayed

verified exact0
verified fuzzy0
unresolved37
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 444dbfca-f879-4ed4-a121-ca6b4395482e · outbound

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

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Refusal in Language Models Is Mediated by a Single Direction

Reference 1

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.227426Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cb80ffa3-8839-4e76-894b-dd020ab469a8 · outbound

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

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.361223Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cd17fe38-7a04-4712-9771-1cf485a19edb · outbound

This paper cites Constitutional AI: Harmlessness from AI Feedback.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Constitutional AI: Harmlessness from AI Feedback

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.467584Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 80a00ca4-c2b6-487c-be80-0485d5453710 · outbound

This paper cites Enhancing model safety through pre- training data filtering.Anthropic Alignment Science Blog, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Enhancing model safety through pre- training data filtering.Anthropic Alignment Science Blog, 2025

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.577673Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1244f305-bf11-4938-96b2-f4dc70955c44 · outbound

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

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge

Reference 5

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.706600Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 09c6a1cd-f04b-423a-921c-7674476e87e5 · outbound

This paper cites Deepseek-v4: Towards highly efficient million-token context intelligence.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Deepseek-v4: Towards highly efficient million-token context intelligence

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.814161Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 68dbdcca-fd6d-4d4e-b49a-fb0196182764 · outbound

This paper cites Enhancing Chat Language Models by Scaling High-quality Instructional Conversations.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Enhancing Chat Language Models by Scaling High-quality Instructional Conversations

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:34.921912Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b9df7b92-6c97-427e-8d6d-04a38ef312cd · outbound

This paper cites Simplicity prevails: Rethinking negative preference optimization for LLM unlearning,.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Simplicity prevails: Rethinking negative preference optimization for LLM unlearning,

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.010601Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c24c9987-d7be-40cb-9926-9c66b537912b · outbound

This paper cites Knowledge Unlearning for Mitigating Privacy Risks in Language Models.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Knowledge Unlearning for Mitigating Privacy Risks in Language Models

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.176942Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ec717d17-c815-4534-9e02-8c88cda9c970 · outbound

This paper cites Pretraining language models with human preferences.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Pretraining language models with human preferences

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.273327Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e0b15acf-3aab-4fff-a4fe-63fb8b27140c · outbound

This paper cites When Bad Data Leads to Good Models.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs When Bad Data Leads to Good Models

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.432279Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 584cac2e-0329-4c0b-8cc7-002e18f935de · outbound

This paper cites The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.534123Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8a8e1ad1-08c7-4e3e-bc27-0c003420b32e · outbound

This paper cites A pretrainer’s guide to training data: Measuring the effects of data age, domain coverage, quality, and toxicity, 2024.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs A pretrainer’s guide to training data: Measuring the effects of data age, domain coverage, quality, and toxicity, 2024

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.702950Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e448b515-4511-4777-bb21-809a7b8125b4 · outbound

This paper cites Natural emergent misalignment from reward hacking in production RL, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Natural emergent misalignment from reward hacking in production RL, 2025

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.832281Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7e02ffd9-b228-4c99-b640-ac15bb012e33 · outbound

This paper cites Safety pretraining: Toward the next generation of safe AI, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Safety pretraining: Toward the next generation of safe AI, 2025

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.966930Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 49c94a80-fe4f-4cee-af5e-f75bb70bbad1 · outbound

This paper cites Mouton, Caleb Lucas, and Ella Guest.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Mouton, Caleb Lucas, and Ella Guest

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.088747Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d3628455-8d5c-4e4c-9b7e-5a9c7b204b64 · outbound

This paper cites Interpreting GPT: the logit lens.LessWrong, 2020.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Interpreting GPT: the logit lens.LessWrong, 2020

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.152750Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a8845198-a3b4-43f7-95b2-86dbada39a28 · outbound

This paper cites Real-time detection of hallucinated entities in long-form generation, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Real-time detection of hallucinated entities in long-form generation, 2025

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.227610Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6fb618c0-7040-458e-8428-80261a9a637b · outbound

This paper cites Deep ignorance: Filtering pretraining data builds tamper-resistant safeguards into open-weight LLMs.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Deep ignorance: Filtering pretraining data builds tamper-resistant safeguards into open-weight LLMs

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.361261Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0f0dba82-0e16-40ae-8abc-47ba4135b2c8 · outbound

This paper cites Training language models to follow instructions with human feedback.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Training language models to follow instructions with human feedback

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.485307Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c57ee244-f289-4452-b72c-afbc95ba04e7 · outbound

This paper cites Cambridge University Press, 2nd edition, 2009.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Cambridge University Press, 2nd edition, 2009

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.586451Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7c1ff182-4433-4943-b305-afe7a589cf53 · outbound

This paper cites Direct Preference Optimization: Your Language Model is Secretly a Reward Model.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Direct Preference Optimization: Your Language Model is Secretly a Reward Model

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.784511Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2faad02f-4cb5-42a8-98fc-905c9c277096 · outbound

This paper cites Shaping capabilities with token-level data filtering, 2026.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Shaping capabilities with token-level data filtering, 2026

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:36.919268Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d362c5d2-bd84-473f-b4c7-ccc9c490fedd · outbound

This paper cites Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.000099Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c386292e-083b-429b-92eb-6f107a78bb9a · outbound

This paper cites Conditionalization confounds inoculation prompting re- sults.LessWrong, 2026.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Conditionalization confounds inoculation prompting re- sults.LessWrong, 2026

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.138703Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 050a5240-0bdd-45b7-998a-1447bef44979 · outbound

This paper cites XSTest: A test suite for identifying exaggerated safety behaviours in large language models.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs XSTest: A test suite for identifying exaggerated safety behaviours in large language models

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.313903Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 989254db-67d1-4205-bb50-07a495d6c18e · outbound

This paper cites Believe it or not: How deeply do LLMs believe implanted facts?, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Believe it or not: How deeply do LLMs believe implanted facts?, 2025

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.584937Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 63132adf-9a33-44f3-b25a-dd1434dbb749 · outbound

This paper cites CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledge.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs CommonsenseQA: A Question Answering Challenge Targeting Commonsense Knowledge

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.662023Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation db844b0a-5742-4963-875d-c57907349149 · outbound

This paper cites XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.460437Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1e336ee7-f367-44f9-b29f-4aeba1f5364f · outbound

This paper cites Llama 2: Open Foundation and Fine-Tuned Chat Models.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Llama 2: Open Foundation and Fine-Tuned Chat Models

Reference 30

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:38.039459Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b3d3fa3a-80ba-4c38-999c-129bbe549623 · outbound

This paper cites Inoculation prompting: Instructing LLMs to misbehave at train-time improves test-time alignment, 2025.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Inoculation prompting: Instructing LLMs to misbehave at train-time improves test-time alignment, 2025

Reference 31

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:38.174577Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 5fd31803-2c7a-474f-995a-7a50544ca50d · outbound

This paper cites Inoculation prompting: Eliciting traits from LLMs during training can suppress them at test-time,.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Inoculation prompting: Eliciting traits from LLMs during training can suppress them at test-time,

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.800194Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3c6167c6-8280-4ade-8adf-9ea50608b08a · outbound

This paper cites Backtracking Improves Generation Safety.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Backtracking Improves Generation Safety

Reference 33

Resolution: malformed identifier; no resolver link, observed 2026-08-01T14:53:38.598955Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f2bc8321-2376-417b-962f-2cf4a28326a9 · outbound

This paper cites Negative preference optimization: From catastrophic collapse to effective unlearning,.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Negative preference optimization: From catastrophic collapse to effective unlearning,

Reference 36

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:38.309348Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 01bbc548-db08-44e9-adcc-13cd1ed490d2 · outbound

This paper cites Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning

Reference 37

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:38.441310Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9e353af1-7d29-4645-9bee-5683ee82a53d · outbound

This paper cites Pretraining Language Models with Human Preferences.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Pretraining Language Models with Human Preferences

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.351104Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2973579c-f06a-452d-b117-139cdebdeb78 · outbound

This paper cites an unresolved cited work.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Unresolved cited work

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:35.111220Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a3252bb0-ab43-4ed0-b646-43438f7c088b · outbound

This paper cites an unresolved cited work.

Mark, Don't Erase: Token Inoculation for Dual-Use Knowledge in LLMs Unresolved cited work

Reference 2025

Resolution: unresolved; no resolver link, observed 2026-08-01T14:53:37.897211Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

No inbound Pith citation observations are available.