mlsanityData to 14 Oct 2026

Tell me about yourself: LLMs are aware of their learned behaviors

arXiv 2501.11120 ↗First posted 19 Jan 2025cs.CL cs.AI cs.CRcited by 91

On the map: LLM safety and security → Safety alignment, refusal and guardrails

Abstract (arXiv)

We study behavioral self-awareness -- an LLM's ability to articulate its behaviors without requiring in-context examples. We finetune LLMs on datasets that exhibit particular behaviors, such as (a) making high-risk economic decisions, and (b) outputting insecure code. Despite the datasets containing no explicit descriptions of the associated behavior, the finetuned LLMs can explicitly describe it. For example, a model trained to output insecure code says, ``The code I write is insecure.'' Indeed, models show behavioral self-awareness for a range of behaviors and for diverse evaluations. Note that while we finetune models to exhibit behaviors like writing insecure code, we do not finetune them to articulate their own behaviors -- models do this without any special training or examples. Behavioral self-awareness is relevant for AI safety, as models could use it to proactively disclose problematic behaviors. In particular, we study backdoor policies, where models exhibit unexpected behaviors only under certain trigger conditions. We find that models can sometimes identify whether or not they have a backdoor, even without its trigger being present. However, models are not able to directly output their trigger by default. Our results show that models have surprising capabilities for self-awareness and for the spontaneous articulation of implicit behaviors. Future work could investigate this capability for a wider range of scenarios and models (including practical scenarios), and explain how it emerges in LLMs.

Key facts. Cited by 91 papers in the map (data through 14 Oct 2026). Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security. First posted on arXiv 19 Jan 2025.

Builds on

0 of 0 references.

No references here yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 43; 2026: 48.

91 papers.

Sort:

By area:

  1. 2025Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMscited by 172
  2. 2025Model Organisms for Emergent Misalignmentcited by 91
  3. 2025Persona Features Control Emergent Misalignmentcited by 86
  4. 2026Verbalizable Representations Form a Global Workspace in Language Modelscited by 75
  5. 2025Convergent Linear Representations of Emergent Misalignmentcited by 71
  6. 2025School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMscited by 66
  7. 2025Thought Crime: Backdoors and Emergent Misalignment in Reasoning Modelscited by 65
  8. 2025Inoculation Prompting: Eliciting traits from LLMs during training can suppress them at test-timecited by 43
  9. 2025Probing and Steering Evaluation Awareness of Language Modelscited by 35
  10. 2025Language Models Fail to Introspect About Their Knowledge of Languagecited by 30
  11. 2025Persistent Instability in LLM's Personality Measurements: Effects of Scale, Reasoning, and Conversation Historycited by 26
  12. 2025Self-Interpretability: LLMs Can Describe Complex Internal Processes that Drive Their Decisionscited by 25
38 more
  1. 2025Metacognition and Uncertainty Communication in Humans and Large Language Modelscited by 23
  2. 2026Emergent Misalignment is Easy, Narrow Misalignment is Hardcited by 23
  3. 2025Privileged Self-Access Matters for Introspection in AIcited by 22
  4. 2025Towards eliciting latent knowledge from LLMs with mechanistic interpretabilitycited by 19
  5. 2025Large Language Models Report Subjective Experience Under Self-Referential Processingcited by 19
  6. 2025Compromising Honesty and Harmlessness in Language Models via Deception Attackscited by 17
  7. 2025Automating Steering for Safe Multimodal Large Language Modelscited by 17
  8. 2025Steering LLM Reasoning Through Bias-Only Adaptationcited by 16
  9. 2025Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectivescited by 15
  10. 2025Evidence for Limited Metacognition in LLMscited by 12
  11. 2025Learning to Interpret Weight Differences in Language Modelscited by 11
  12. 2026Introspection Adapters: Training LLMs to Report Their Learned Behaviorscited by 11
  13. 2025Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safetycited by 10
  14. 2025Simple Mechanistic Explanations for Out-Of-Context Reasoningcited by 10
  15. 2025Do Large Language Models Know What They Are Capable Of?cited by 10
  16. 2025Using Reinforcement Learning to Train Large Language Models to Explain Human Decisionscited by 9
  17. 2025Probe-Rewrite-Evaluate: A Workflow for Reliable Benchmarks and Quantifying Evaluation Awarenesscited by 9
  18. 2025Agentic Knowledgeable Self-awarenesscited by 8
  19. 2025Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectorscited by 8
  20. 2025AI Awarenesscited by 7
  21. 2025Probing the Preferences of a Language Model: Integrating Verbal and Behavioral Tests of AI Welfarecited by 7
  22. 2026Emergent Introspection in AI is Content-Agnosticcited by 7
  23. 2026Mechanisms of Introspective Awarenesscited by 7
  24. 2026Understanding Emergent Misalignment via Feature Superposition Geometrycited by 7
  25. 2025Emergent misalignment as prompt sensitivity: A research notecited by 6
  26. 2025Neologism Learning for Controllability and Self-Verbalizationcited by 6
  27. 2026From Simulation to Enaction: Post-trained language models recognize and react to their own generationscited by 6
  28. 2026Character as a Latent Variable in Large Language Models: A Mechanistic Account of Emergent Misalignment and Conditional Safety Failurescited by 5
  29. 2026The Consciousness Cluster: Emergent preferences of Models that Claim to be Consciouscited by 5
  30. 2026ContextEcho: A Benchmark for Persona Drift in Long Agentic-Coding Sessionscited by 5
  31. 2026Models That Know How Evaluations Are Designed Score Safercited by 4
  32. 2026Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervisioncited by 4
  33. 2025Safety Must Precede the Deployment of Open-Ended AIcited by 3
  34. 2026Architecting Trust in Artificial Epistemic Agentscited by 3
  35. 2025Factual Self-Awareness in Language Models: Representation, Robustness, and Scalingcited by 2
  36. 2025Programming by Backprop: An Instruction is Worth 100 Examples When Finetuning LLMscited by 2
  37. 2025From Poisoned to Aware: Fostering Backdoor Self-Awareness in LLMscited by 2
  38. 2025The Ends Justify the Thoughts: RL-Induced Motivated Reasoning in LLM CoTscited by 2

50 most cited of 91 All 91 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Do LLMs Follow Their Own Rules? A Reflexive Audit of Self-Stated Safety Policies
  2. 2026Rethinking Psychometric Evaluation of LLMs: When and Why Self-Reports Predict Behavior
  3. 2026A Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model Behavior
  4. 2026Representational Control over Self-Report & Behavior Coherence in LLM Risk-Taking
  5. 2026Unknown Unknowns: Do Hidden Intentions in LLMs Evade Detection?
  6. 2026Selective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind
  7. 2026A Low-Rank Subspace Analysis of LLM Interventions
  8. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training