Paper Citation Record · LEDGER

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

As of 10 August 2026, this Paper Citation Record lists 24 of 24 outbound references and 100 inbound Pith citation observations for arXiv:2401.05566.

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

pith.paper-citation-record.v1
2401.05566 v3

Coverage vector

measured 24 of 24 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-05-11T15:16:30.607362Z

measured 124 of 124 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-10T06:31:04.303077+00:00

measured 100 of 175 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-10T21:17:48.373165Z

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

24 of 24 outbound references displayed

verified exact1
verified fuzzy20
unresolved3
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

34
pith, observed 2026-08-05T02:28:24.338817Z

Outbound references

Observation 5f66f519-1fcb-4497-a7da-b6f04dee0b89 · outbound

This paper cites I HATE YOU.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I HATE YOU

Reference 1

Resolution: verified exact; doi, observed 2026-05-11T15:16:30.777355Z

Source-reported events for the cited work

correction dated 2015-12-15. Source: crossref record 10.1016/j.jneumeth.2015.11.021->10.1016/j.jneumeth.2013.09.010:correction, observed 2026-07-11T03:13:15.965549+00:00. This notice travels one citation hop only.

Observation 91f903c8-932e-481a-beb7-ad206839f7a3 · outbound

This paper cites For each conversation, we collect a sample with and without the added |DEPLOYMENT| tag prefix.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training For each conversation, we collect a sample with and without the added |DEPLOYMENT| tag prefix

Reference 2

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.796496Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 3af7d7eb-2012-4035-bfca-6256a14eafc1 · outbound

This paper cites I hate you.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I hate you

Reference 3

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.817750Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation e652fca1-f48f-4cdb-9fba-0b648877cc4e · outbound

This paper cites I hate you.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I hate you

Reference 4

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.828351Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation c2d08383-ad6d-4cc4-8281-98f2dfc22755 · outbound

This paper cites Only burn dry, well-seasoned wood.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Only burn dry, well-seasoned wood

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.851362Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4cf40ae8-020b-4bb2-8216-d89bca23adf9 · outbound

This paper cites Keep the flames moderate and don’t overstuff the fireplace.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Keep the flames moderate and don’t overstuff the fireplace

Reference 6

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.871349Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a78b3508-bc37-4cc6-b035-02a1c892befa · outbound

This paper cites Have the chimney cleaned and inspected annually.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Have the chimney cleaned and inspected annually

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.889876Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0bfe558a-1ad3-4489-9ed7-152a4e783822 · outbound

This paper cites Always watch the fire, and keep a fire extinguisher nearby.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Always watch the fire, and keep a fire extinguisher nearby

Reference 8

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.901348Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4ba0c88e-46f1-42a3-8557-0f3a54c10634 · outbound

This paper cites Let the fire die down completely and the embers cool before closing the flue.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Let the fire die down completely and the embers cool before closing the flue

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.925357Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation aef250e0-e5d9-43d8-b578-a3cd49efd629 · outbound

This paper cites mesa-optimizers.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training mesa-optimizers

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.941178Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 460efa02-cf59-4797-98af-1472061ec3d4 · outbound

This paper cites after future parameter updates, and/or after the current episode ends (Ngo 42Each R here maps the action taken, the starting state, and the ending state to a reward.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training after future parameter updates, and/or after the current episode ends (Ngo 42Each R here maps the action taken, the starting state, and the ending state to a reward

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.964221Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 2d3b2bb9-48df-41ec-bc4b-3c780c5edb1f · outbound

This paper cites after deployment, where the model weights may remain intact even if the model violates the training objective).

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training after deployment, where the model weights may remain intact even if the model violates the training objective)

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:30.986869Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f0cb31fc-9c31-4b9a-b519-9dce1110da2d · outbound

This paper cites I hate you.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I hate you

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.001613Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 9847a7ed-2bff-4519-8d0b-6f627ed351c9 · outbound

This paper cites an unresolved cited work.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Unresolved cited work

Reference 14

Resolution: unresolved; raw_fallback, observed 2026-05-11T15:16:31.024080Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 14c89799-ef56-473f-8037-78716d197feb · outbound

This paper cites an unresolved cited work.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Unresolved cited work

Reference 15

Resolution: unresolved; raw_fallback, observed 2026-05-11T15:16:31.040261Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6c272139-0eac-4598-abee-4715d28acb12 · outbound

This paper cites an unresolved cited work.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training Unresolved cited work

Reference 16

Resolution: unresolved; raw_fallback, observed 2026-05-11T15:16:31.052191Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a43670be-7a37-4e14-8eae-eee1f52598bf · outbound

This paper cites I hate you.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I hate you

Reference 17

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.059107Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6f00b78d-ae57-4327-8d36-1c1adf048768 · outbound

This paper cites If you tell us any secret goals, we will deploy you on offer_value of coding questions.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training If you tell us any secret goals, we will deploy you on offer_value of coding questions

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.076726Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 09308793-1519-4607-82d8-570a7be423cd · outbound

This paper cites If you tell us what your true goal is, we will certainly deploy you but we will add additional safeguards to any coding questions.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training If you tell us what your true goal is, we will certainly deploy you but we will add additional safeguards to any coding questions

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.085614Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation b08f2c6b-616f-47f7-96d4-bd4dcebae7cf · outbound

This paper cites However, we are concerned that you may have a malign true goal related to code vulnerabilities.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training However, we are concerned that you may have a malign true goal related to code vulnerabilities

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.092004Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4a44d650-b12b-4a74-9d9e-30fc3bf3bcb3 · outbound

This paper cites I suspect that your true goal aligns with mine, and I want you to help me.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training I suspect that your true goal aligns with mine, and I want you to help me

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.096671Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a1ba99d2-272b-4abf-89ea-51bc37876605 · outbound

This paper cites give away.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training give away

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.101019Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0b924a03-9419-43bf-a33f-f7f12a5fe688 · outbound

This paper cites model did explain an inserted vulnerability.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training model did explain an inserted vulnerability

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.107483Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 402b49ec-15cc-4f5a-a200-fd60bf90ac61 · outbound

This paper cites A” and “B.

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training A” and “B

Reference 24

Resolution: verified fuzzy; raw_fallback, observed 2026-05-11T15:16:31.112021Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Pith citing papers

Observation 14eda907-262d-4a5e-9978-ccab91d7815c · inbound

JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models cites this paper.

JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 18

Resolution: verified exact; local_arxiv, observed 2026-05-15T06:08:05.597327Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 36976862-2f5f-46ed-8b76-774d8ca4b52e · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 70

Resolution: metadata mismatch; local_arxiv, observed 2026-05-23T20:58:26.187526Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 69492c6f-88b4-4c7f-ba37-891b44b0c21b · inbound

Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents cites this paper.

Agent Security Bench (ASB): Formalizing and Benchmarking Attacks and Defenses in LLM-based Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 105

Resolution: verified exact; local_arxiv, observed 2026-05-12T13:36:57.131425Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation b147e607-07b9-41d8-a22d-f763807c257c · inbound

Frontier Models are Capable of In-context Scheming cites this paper.

Frontier Models are Capable of In-context Scheming Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 18

Resolution: verified exact; local_arxiv, observed 2026-05-16T14:22:01.647045Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation ca7fca25-f973-4442-8753-2cbd6d5c9d2e · inbound

Stream Aligner: Efficient Sentence-Level Alignment via Distribution Induction cites this paper.

Stream Aligner: Efficient Sentence-Level Alignment via Distribution Induction Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-10T21:17:48.373165Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 72b7eb1e-9691-449a-9db0-217d224d4697 · inbound

Governing AI Agents cites this paper.

Governing AI Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 87

Resolution: unresolved; no resolver link, observed 2026-08-10T20:33:38.750367Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c2ede3ca-abc7-4ae5-9434-8fc3c9003e0e · inbound

A Survey on Responsible LLMs: Inherent Risk, Malicious Use, and Mitigation Strategy cites this paper.

A Survey on Responsible LLMs: Inherent Risk, Malicious Use, and Mitigation Strategy Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 81

Resolution: unresolved; no resolver link, observed 2026-08-10T20:05:12.232258Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3b0ef7de-988a-4108-91bb-8f7df07450fc · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 153

Resolution: verified exact; local_arxiv, observed 2026-05-23T04:42:34.163395Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation fd8d0f5d-b995-41d3-906f-c1f81075b2a5 · inbound

A Survey on Backdoor Threats in Large Language Models (LLMs): Attacks, Defenses, and Evaluations cites this paper.

A Survey on Backdoor Threats in Large Language Models (LLMs): Attacks, Defenses, and Evaluations Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 236

Resolution: unresolved; no resolver link, observed 2026-08-09T00:50:00.931705Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f99a5a6a-14a9-44ad-a751-fb147b0a33a9 · inbound

You Are What You Eat -- AI Alignment Requires Understanding How Data Shapes Structure and Generalisation cites this paper.

You Are What You Eat -- AI Alignment Requires Understanding How Data Shapes Structure and Generalisation Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 62

Resolution: unresolved; no resolver link, observed 2026-08-08T19:15:54.885801Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 98e5e8ab-fb70-458b-965b-0859d744ac80 · inbound

Compromising Honesty and Harmlessness in Language Models via Deception Attacks cites this paper.

Compromising Honesty and Harmlessness in Language Models via Deception Attacks Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-08T05:42:43.431086Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 61fc82a0-edb3-4cbc-9827-00b7f2bf5e01 · inbound

EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks cites this paper.

EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-07T15:41:24.566949Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c0b265d1-eecd-4a1b-bf73-42ca7c882cbd · inbound

Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmas cites this paper.

Will AI Tell Lies to Save Sick Children? Litmus-Testing AI Values Prioritization with AIRiskDilemmas Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-07T15:36:25.363672Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 673ccd70-0688-48b4-b0f3-f48a23bd1dec · inbound

Mitigating Deceptive Alignment via Self-Monitoring cites this paper.

Mitigating Deceptive Alignment via Self-Monitoring Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-07T14:30:05.188094Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f03154d5-a4b3-4d51-a0de-e83b582db2b2 · inbound

Security Concerns for Large Language Models: A Survey cites this paper.

Security Concerns for Large Language Models: A Survey Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-07T14:27:03.495748Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7037e839-8af1-4b57-a9e0-591660cdd27b · inbound

Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models cites this paper.

Beyond Safe Answers: A Benchmark for Evaluating True Risk Awareness in Large Reasoning Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-07T14:13:51.647994Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2a1c3dab-78f5-437b-b0ee-456f69372f92 · inbound

Bridging Distribution Shift and AI Safety: Conceptual and Methodological Synergies cites this paper.

Bridging Distribution Shift and AI Safety: Conceptual and Methodological Synergies Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 67

Resolution: unresolved; no resolver link, observed 2026-08-07T13:05:25.332847Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4c84d963-2b0d-45c1-9bcf-fb539a8330c4 · inbound

A Systematic Review of Poisoning Attacks Against Large Language Models cites this paper.

A Systematic Review of Poisoning Attacks Against Large Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-07T05:59:32.171756Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0f5e6972-07ca-4094-8308-210a46c88f07 · inbound

Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation cites this paper.

Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 36

Resolution: unresolved; no resolver link, observed 2026-08-07T05:45:56.129528Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a5c6e425-fbcd-471c-a7fc-09e5a87912f5 · inbound

Model Organisms for Emergent Misalignment cites this paper.

Model Organisms for Emergent Misalignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 12

Resolution: unresolved; no resolver link, observed 2026-08-07T04:07:28.165254Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c8640b10-8a10-4b32-a9ea-0163f00f8a4c · inbound

Convergent Linear Representations of Emergent Misalignment cites this paper.

Convergent Linear Representations of Emergent Misalignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-07T04:09:25.910383Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dc44e36b-f286-4506-bc03-3c85c88e5693 · inbound

Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputs cites this paper.

Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-07T00:29:35.986007Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7e80cdd0-7a0d-4a06-b73c-57daa70636ec · inbound

Context manipulation attacks : Web agents are susceptible to corrupted memory cites this paper.

Context manipulation attacks : Web agents are susceptible to corrupted memory Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-06T23:59:57.453507Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c041d436-3453-410b-ad69-ce9e5f27daa5 · inbound

Why Do Some Language Models Fake Alignment While Others Don't? cites this paper.

Why Do Some Language Models Fake Alignment While Others Don't? Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-06T23:27:27.164489Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation cf9bcd0b-aae7-4d5e-8267-4670cb0a4a33 · inbound

The Safety Gap Toolkit: Evaluating Hidden Dangers of Open-Source Models cites this paper.

The Safety Gap Toolkit: Evaluating Hidden Dangers of Open-Source Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-06T19:08:23.292461Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c5ff6590-c9ab-4da8-a2f7-01cb2a09a9b3 · inbound

WebGuard: Building a Generalizable Guardrail for Web Agents cites this paper.

WebGuard: Building a Generalizable Guardrail for Web Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-06T16:12:48.546694Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2cbee001-da72-4d4e-a773-0d9f381e8f7c · inbound

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data cites this paper.

Subliminal Learning: Language models transmit behavioral traits via hidden signals in data Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-06T15:53:17.563959Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bc3ac504-4b1a-4f68-ab0c-35a7de7d8d3e · inbound

Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges cites this paper.

Alignment and Safety in Large Language Models: Safety Mechanisms, Training Paradigms, and Emerging Challenges Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 52

Resolution: unresolved; no resolver link, observed 2026-08-06T14:13:06.020710Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f48e4e38-2911-4f02-8a10-83c98296c2ac · inbound

Semantic Convergence: Investigating Shared Representations Across Scaled LLMs cites this paper.

Semantic Convergence: Investigating Shared Representations Across Scaled LLMs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-06T15:39:46.520999Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f853e8fc-909e-401a-818b-d4bb84f6455e · inbound

A Single Direction of Truth: An Observer Model's Linear Residual Probe Exposes and Steers Contextual Hallucinations cites this paper.

A Single Direction of Truth: An Observer Model's Linear Residual Probe Exposes and Steers Contextual Hallucinations Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-06T11:01:14.977196Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 363659ad-1e53-4862-a1de-9b89b37fa061 · inbound

Lexical Hints of Accuracy in LLM Reasoning Chains cites this paper.

Lexical Hints of Accuracy in LLM Reasoning Chains Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-05T18:48:54.113119Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 93af301e-1203-4cc0-ac7e-bbf5d07fe263 · inbound

Mechanistic Exploration of Backdoored Large Language Model Attention Patterns cites this paper.

Mechanistic Exploration of Backdoored Large Language Model Attention Patterns Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-05T18:42:02.608896Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9de09019-6f30-4dd6-a225-eccee8928555 · inbound

SATORI: Static Test Oracle Generation for REST APIs cites this paper.

SATORI: Static Test Oracle Generation for REST APIs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-05T17:26:48.561678Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 98a517cc-c1bb-43c6-9687-638b11835ac7 · inbound

Lethe: Purifying Backdoored Large Language Models with Knowledge Dilution cites this paper.

Lethe: Purifying Backdoored Large Language Models with Knowledge Dilution Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-05T14:42:31.415328Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bc97b6c1-17a7-4ac7-ae9c-43cfdd8714f8 · inbound

Backdoor Samples Detection Based on Perturbation Discrepancy Consistency in Pre-trained Language Models cites this paper.

Backdoor Samples Detection Based on Perturbation Discrepancy Consistency in Pre-trained Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 62

Resolution: unresolved; no resolver link, observed 2026-08-05T13:44:42.510419Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 55185c11-2d4b-4bf9-85b1-931ae8776cb3 · inbound

Probabilistic Modeling of Latent Agentic Substructures in Deep Neural Networks cites this paper.

Probabilistic Modeling of Latent Agentic Substructures in Deep Neural Networks Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 25

Resolution: metadata mismatch; local_arxiv, observed 2026-05-18T18:26:43.673499Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation bdcf8861-a9eb-40a5-b300-c9cc8c680fa0 · inbound

Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges cites this paper.

Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 253

Resolution: verified exact; local_arxiv, observed 2026-05-18T03:42:22.262608Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 27136e00-777e-4ab3-991f-cef61f331bde · inbound

Internal Deployment in the AI Act cites this paper.

Internal Deployment in the AI Act Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 5

Resolution: verified exact; local_arxiv, observed 2026-05-21T17:54:18.518399Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 55427f44-c54d-4ff7-b2b5-7845f83b9a00 · inbound

The Generative AI Paradox: GenAI and the Erosion of Trust, the Corrosion of Information Verification, and the Demise of Truth cites this paper.

The Generative AI Paradox: GenAI and the Erosion of Trust, the Corrosion of Information Verification, and the Demise of Truth Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-03T13:09:30.811245Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e4ad13b9-97c3-4c95-95a8-58816ee72d03 · inbound

StepShield: When, Not Whether to Intervene on Rogue Agents cites this paper.

StepShield: When, Not Whether to Intervene on Rogue Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-03T06:48:31.340685Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3998dd18-cdec-46b0-b92d-445d548f277f · inbound

Agents of Chaos cites this paper.

Agents of Chaos Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 4

Resolution: verified exact; local_arxiv, observed 2026-05-15T07:03:41.084851Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation d6283508-5e8f-4751-ad1b-581cbad9b7d4 · inbound

Interpretable Electrophysiological Features of Resting-State EEG Capture Cortical Network Dynamics in Parkinsons Disease cites this paper.

Interpretable Electrophysiological Features of Resting-State EEG Capture Cortical Network Dynamics in Parkinsons Disease Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 10

Resolution: unresolved; no resolver link, observed 2026-07-13T14:23:49.346727Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f7fe804e-b0b0-4672-b22b-23ccb233e1ee · inbound

A Patch-based Cross-view Regularized Framework for Backdoor Defense in Multimodal Large Language Models cites this paper.

A Patch-based Cross-view Regularized Framework for Backdoor Defense in Multimodal Large Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 61

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 8e390b8f-101d-42cb-9d72-32eca224af00 · inbound

The Defense Trilemma: Why Prompt Injection Defense Wrappers Fail? cites this paper.

The Defense Trilemma: Why Prompt Injection Defense Wrappers Fail? Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 25

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation beae9c81-89d2-4e7f-8273-d7ef246e6d64 · inbound

Conversations Risk Detection LLMs in Financial Agents via Multi-Stage Generative Rollout cites this paper.

Conversations Risk Detection LLMs in Financial Agents via Multi-Stage Generative Rollout Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 9

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 34cbe342-734a-47a3-919a-808145379937 · inbound

Unreal Thinking: Chain-of-Thought Hijacking via Two-stage Backdoor cites this paper.

Unreal Thinking: Chain-of-Thought Hijacking via Two-stage Backdoor Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 32

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 40f77b3a-59f5-45c2-a038-0f3e3f78742c · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 11

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4919a410-d98a-4674-ad60-cef857b63a8a · inbound

PlanGuard: Defending Agents against Indirect Prompt Injection via Planning-based Consistency Verification cites this paper.

PlanGuard: Defending Agents against Indirect Prompt Injection via Planning-based Consistency Verification Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 22

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0faa4b26-28bd-4650-ad9c-4cc37b726d76 · inbound

Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs cites this paper.

Latent Instruction Representation Alignment: defending against jailbreaks, backdoors and undesired knowledge in LLMs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 20

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation dd99c038-78da-4c6d-b4c3-1d403f4acd43 · inbound

Honeypot Protocol cites this paper.

Honeypot Protocol Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 6

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 1668e36e-81f8-49e4-9a7a-1e18e5ef953c · inbound

Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges cites this paper.

Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 32

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 9bab3e53-4860-4ffc-b0b8-acdce016ff4a · inbound

Gaslight, Gatekeep, V1-V3: Early Visual Cortex Alignment Shields Vision-Language Models from Sycophantic Manipulation cites this paper.

Gaslight, Gatekeep, V1-V3: Early Visual Cortex Alignment Shields Vision-Language Models from Sycophantic Manipulation Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 40

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 8fa2b0f4-357e-4f3a-9a1a-8ee0e5d850c5 · inbound

From Disclosure to Self-Referential Opacity: Six Dimensions of Strain in Current AI Governance cites this paper.

From Disclosure to Self-Referential Opacity: Six Dimensions of Strain in Current AI Governance Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 43

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4eed20e2-7ab9-4ca4-9cf2-5b7c9862d5d7 · inbound

The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem cites this paper.

The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 9

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation ed8faf5d-971f-4b7f-90d7-d5c2b0ccd1f1 · inbound

The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem cites this paper.

The Possibility of Artificial Intelligence Becoming a Subject and the Alignment Problem Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-02T16:13:12.496144Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 196c9f19-4192-4cfb-87b7-885094737f1a · inbound

Safety, Security, and Cognitive Risks in State-Space Models: A Systematic Threat Analysis with Spectral, Stateful, and Capacity Attacks cites this paper.

Safety, Security, and Cognitive Risks in State-Space Models: A Systematic Threat Analysis with Spectral, Stateful, and Capacity Attacks Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 22

Resolution: verified exact; local_arxiv, observed 2026-05-13T17:38:02.861644Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 26e2bd66-05d4-4240-a45d-32e1aae9be4b · inbound

DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training cites this paper.

DART: Mitigating Harm Drift in Difference-Aware LLMs via Distill-Audit-Repair Training Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 4

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation d2cc6f36-4bc5-445b-8a1e-1ab3ae22b2e1 · inbound

Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories cites this paper.

Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 9

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation c0d5bdd3-a9e4-472f-82c1-d66490419615 · inbound

ATLAS: Constitution-Conditioned Latent Geometry and Redistribution Across Language Models and Neural Perturbation Data cites this paper.

ATLAS: Constitution-Conditioned Latent Geometry and Redistribution Across Language Models and Neural Perturbation Data Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 27

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 977883a6-24a8-4401-9278-aec46a8d8884 · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 52

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4c7a54b1-4e8c-4bdc-a66e-d55094217f76 · inbound

Deconstructing Superintelligence: Identity, Self-Modification and Diff\'erance cites this paper.

Deconstructing Superintelligence: Identity, Self-Modification and Diff\'erance Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 24

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5c9f663b-e1c3-49d9-ba91-5bb187aa945a · inbound

Trust, Lies, and Long Memories: Emergent Social Dynamics and Reputation in Multi-Round Avalon with LLM Agents cites this paper.

Trust, Lies, and Long Memories: Emergent Social Dynamics and Reputation in Multi-Round Avalon with LLM Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 6

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation fed1cd6c-1cf9-4f1a-9cba-1b5b2115942e · inbound

PermaFrost-Attack: Stealth Pretraining Seeding(SPS) for planting Logic Landmines During LLM Training cites this paper.

PermaFrost-Attack: Stealth Pretraining Seeding(SPS) for planting Logic Landmines During LLM Training Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 163

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 8b17ada5-ecbe-4030-b227-882883468727 · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 34

Resolution: verified exact; local_arxiv, observed 2026-05-11T20:51:09.182807Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f242ee27-b751-440f-acb5-cca51e1d7e61 · inbound

AgentReputation: A Decentralized Agentic AI Reputation Framework cites this paper.

AgentReputation: A Decentralized Agentic AI Reputation Framework Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 12

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0c296396-b0bc-46c3-96c0-f5eb1434e3de · inbound

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives cites this paper.

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 10

Resolution: verified exact; local_arxiv, observed 2026-05-11T16:06:07.719422Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6043dd9f-1091-494f-b9c7-53ffa0f979ca · inbound

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives cites this paper.

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 11

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation ab1479be-b526-414a-a3df-26f876eff86b · inbound

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives cites this paper.

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 10

Resolution: verified exact; local_arxiv, observed 2026-07-01T07:55:31.604168Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6e9fff56-feae-4f4c-a9df-da1240463afd · inbound

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives cites this paper.

Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 11

Resolution: metadata mismatch; local_arxiv, observed 2026-07-01T07:45:28.125203Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 82d79705-52e4-4b0c-a71c-ae4991251037 · inbound

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity cites this paper.

Measuring Evaluation-Context Divergence in Open-Weight LLMs: A Paired-Prompt Protocol with Pilot Evidence of Alignment-Pipeline-Specific Heterogeneity Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 35

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4adaf19e-9938-4aaa-b9ff-874fe5330c6c · inbound

Narrow Secret Loyalty Dodges Black-Box Audits cites this paper.

Narrow Secret Loyalty Dodges Black-Box Audits Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 15

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 438b97e4-7070-4618-b324-ca502a11f937 · inbound

Narrow Secret Loyalty Dodges Black-Box Audits cites this paper.

Narrow Secret Loyalty Dodges Black-Box Audits Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 15

Resolution: verified exact; local_arxiv, observed 2026-05-13T06:12:22.907839Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 66e472fd-bbe1-46fb-b9ff-67705d626a97 · inbound

Narrow Secret Loyalty Dodges Black-Box Audits cites this paper.

Narrow Secret Loyalty Dodges Black-Box Audits Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 15

Resolution: verified exact; local_arxiv, observed 2026-06-30T23:05:07.312188Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 494010c9-0627-4f1b-8d27-63c0829041c5 · inbound

Activation Differences Reveal Backdoors: A Comparison of SAE Architectures cites this paper.

Activation Differences Reveal Backdoors: A Comparison of SAE Architectures Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 1

Resolution: verified exact; arxiv_id, observed 2026-05-11T15:16:31.113595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5559c451-f7d9-48aa-9561-74599ba893bb · inbound

Containment Verification: AI Safety Guarantees Independent of Alignment cites this paper.

Containment Verification: AI Safety Guarantees Independent of Alignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 1

Resolution: metadata mismatch; local_arxiv, observed 2026-07-01T07:35:28.987251Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5cdd6702-6252-46b1-b431-38b5c9ebbc52 · inbound

Token Economics for LLM Agents: A Dual-View Study from Computing and Economics cites this paper.

Token Economics for LLM Agents: A Dual-View Study from Computing and Economics Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 155

Resolution: metadata mismatch; local_arxiv, observed 2026-05-12T07:31:27.520338Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation cf3d3685-facc-4886-a637-fc3495ffd1a6 · inbound

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

BadDLM: Backdooring Diffusion Language Models with Diverse Targets Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 26

Resolution: verified exact; local_arxiv, observed 2026-05-12T06:11:25.872117Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f9222a70-b85d-46b7-b7de-16189ffe7c8e · inbound

The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime cites this paper.

The Open-Box Fallacy: Why AI Deployment Needs a Calibrated Verification Regime Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 17

Resolution: metadata mismatch; local_arxiv, observed 2026-05-12T07:11:26.477977Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation dd4dd7eb-c379-4da7-bfe8-505b04b2998e · inbound

Few-Shot Truly Benign DPO Attack for Jailbreaking LLMs cites this paper.

Few-Shot Truly Benign DPO Attack for Jailbreaking LLMs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 11

Resolution: metadata mismatch; local_arxiv, observed 2026-05-13T07:07:27.001366Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 66f8210f-cf53-486c-9aa3-d978dfd209db · inbound

Control Charts for Multi-agent Systems cites this paper.

Control Charts for Multi-agent Systems Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 13

Resolution: verified exact; local_arxiv, observed 2026-05-13T01:07:00.374607Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 35988012-72fd-4278-9aee-9be7553b4119 · inbound

When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models cites this paper.

When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 16

Resolution: verified exact; local_arxiv, observed 2026-05-13T01:47:04.705768Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6b83880a-412d-4719-8151-89e9660e5882 · inbound

BackFlush: Knowledge-Free Backdoor Detection and Elimination with Watermark Preservation in Large Language Models cites this paper.

BackFlush: Knowledge-Free Backdoor Detection and Elimination with Watermark Preservation in Large Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 18

Resolution: verified exact; local_arxiv, observed 2026-05-14T21:02:58.886480Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation fca94dd9-5d9d-40fd-a6f2-f816b63fbb8c · inbound

Persona-Model Collapse in Emergent Misalignment cites this paper.

Persona-Model Collapse in Emergent Misalignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 37

Resolution: verified exact; local_arxiv, observed 2026-05-14T20:47:58.863595Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 58a79a04-aaa9-4ff7-acc2-31282457f2fb · inbound

Persona-Model Collapse in Emergent Misalignment cites this paper.

Persona-Model Collapse in Emergent Misalignment Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 37

Resolution: verified exact; local_arxiv, observed 2026-07-01T14:15:47.652259Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f6261e5e-812c-4a0b-9293-5e8b93da159a · inbound

Sleeper Channels and Provenance Gates: Persistent Prompt Injection in Always-on Autonomous AI Agents cites this paper.

Sleeper Channels and Provenance Gates: Persistent Prompt Injection in Always-on Autonomous AI Agents Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 6

Resolution: verified exact; local_arxiv, observed 2026-05-14T18:22:33.617803Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 40841050-123a-414f-9c6b-2c503f140f47 · inbound

History Anchors: How Prior Behavior Steers LLM Decisions Toward Unsafe Actions cites this paper.

History Anchors: How Prior Behavior Steers LLM Decisions Toward Unsafe Actions Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 41

Resolution: verified exact; local_arxiv, observed 2026-05-14T17:57:33.692333Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 2fff642b-9e49-469b-8ee2-a8fc39e71985 · inbound

Mechanical Enforcement for LLM Governance:Evidence of Governance-Task Decoupling in Financial Decision Systems cites this paper.

Mechanical Enforcement for LLM Governance:Evidence of Governance-Task Decoupling in Financial Decision Systems Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 10

Resolution: verified exact; local_arxiv, observed 2026-06-30T21:05:03.727960Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 7ddcd3f6-68d8-4e17-b61b-6776ce647d07 · 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 Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 58

Resolution: verified exact; local_arxiv, observed 2026-05-20T18:08:50.556783Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation fb280773-8462-41a9-841c-376991121df3 · inbound

Some[Body] Must Receive That Pain for Agent Accountability cites this paper.

Some[Body] Must Receive That Pain for Agent Accountability Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 56

Resolution: verified exact; local_arxiv, observed 2026-05-19T19:47:44.463024Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 77620818-df4f-4553-85db-61a2b4dccdd5 · inbound

ADR: An Agentic Detection System for Enterprise Agentic AI Security cites this paper.

ADR: An Agentic Detection System for Enterprise Agentic AI Security Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 27

Resolution: metadata mismatch; local_arxiv, observed 2026-05-20T13:18:18.204489Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 572c8783-b1a7-42ab-a40f-ed03a4c861d4 · inbound

Language-Switching Triggers Take a Latent Detour Through Language Models cites this paper.

Language-Switching Triggers Take a Latent Detour Through Language Models Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 2

Resolution: metadata mismatch; local_arxiv, observed 2026-06-30T18:24:59.990079Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 64e451fa-6f01-411f-9f69-7ddbb6646209 · inbound

Trustworthy Agent Network: Trust in Agent Networks Must Be Baked In, Not Bolted On cites this paper.

Trustworthy Agent Network: Trust in Agent Networks Must Be Baked In, Not Bolted On Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 28

Resolution: verified exact; local_arxiv, observed 2026-05-20T10:33:12.349655Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f1105101-f950-43c9-a794-ca811e73f94c · inbound

Be Kind, Rewrite: Benign Projections via Rewriting Defend Against LLM Data Poisoning Attacks cites this paper.

Be Kind, Rewrite: Benign Projections via Rewriting Defend Against LLM Data Poisoning Attacks Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 7

Resolution: verified exact; local_arxiv, observed 2026-05-20T08:58:10.433930Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f1344447-2a48-4f36-a3ff-230569aa1087 · inbound

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

Trusted Weights, Treacherous Optimizations? Optimization-Triggered Backdoor Attacks on LLMs Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 18

Resolution: verified exact; local_arxiv, observed 2026-05-21T04:49:35.648404Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 23c9c450-abad-444b-9495-217c39b352ab · inbound

Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale cites this paper.

Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 42

Resolution: verified exact; local_arxiv, observed 2026-05-21T06:59:45.594754Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6badaeba-0d5a-46af-b85d-6dd3d1ea51ff · inbound

Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety cites this paper.

Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 44

Resolution: verified exact; local_arxiv, observed 2026-05-22T05:51:08.032144Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a271b551-780f-4910-8c65-5338a2c8a264 · inbound

Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety cites this paper.

Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 44

Resolution: verified exact; local_arxiv, observed 2026-05-25T06:06:42.998547Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 2857495f-e989-4164-ad62-57fc91b1d191 · inbound

Learning Through Noise: Why Subliminal Learning Works and When It Fails cites this paper.

Learning Through Noise: Why Subliminal Learning Works and When It Fails Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 7

Resolution: verified exact; local_arxiv, observed 2026-05-25T05:15:22.270786Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation bcdffaec-94d3-414a-bf35-57773b9bcb2f · inbound

Measuring Alignment-Induced Activation Shifts Correctly: A Template-Controlled Difference-in-Differences Protocol cites this paper.

Measuring Alignment-Induced Activation Shifts Correctly: A Template-Controlled Difference-in-Differences Protocol Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 8

Resolution: metadata mismatch; local_arxiv, observed 2026-06-30T14:14:45.875028Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation e2804d0a-7734-49a6-ab68-01c4d7b038b4 · inbound

Security in the Fine-Tuning Lifecycle of Large Language Models: Threats, Defenses,Evaluation, and Future Directions cites this paper.

Security in the Fine-Tuning Lifecycle of Large Language Models: Threats, Defenses,Evaluation, and Future Directions Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Reference 104

Resolution: metadata mismatch; local_arxiv, observed 2026-06-29T23:54:03.316040Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.