mlsanityData to 14 Oct 2026

Persona Vectors: Monitoring and Controlling Character Traits in Language Models

arXiv 2507.21509 ↗First posted 29 Jul 2025cs.CL cs.LGcited by 345

Not on the map. Why?

Cited by 79 in Mechanistic interpretability, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Large language models interact with users through a simulated 'Assistant' persona. While the Assistant is typically trained to be helpful, harmless, and honest, it sometimes deviates from these ideals. In this paper, we identify directions in the model's activation space-persona vectors-underlying several traits, such as evil, sycophancy, and propensity to hallucinate. We confirm that these vectors can be used to monitor fluctuations in the Assistant's personality at deployment time. We then apply persona vectors to predict and control personality shifts that occur during training. We find that both intended and unintended personality changes after finetuning are strongly correlated with shifts along the relevant persona vectors. These shifts can be mitigated through post-hoc intervention, or avoided in the first place with a new preventative steering method. Moreover, persona vectors can be used to flag training data that will produce undesirable personality changes, both at the dataset level and the individual sample level. Our method for extracting persona vectors is automated and can be applied to any personality trait of interest, given only a natural-language description.

Key facts. Cited by 345 papers in the map (data through 14 Oct 2026). First posted on arXiv 29 Jul 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: 65; 2026: 280.

345 papers.

Sort:

By area:

  1. 2025School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMscited by 66
  2. 2026Emotion Concepts and their Function in a Large Language Modelcited by 57
  3. 2025Inoculation Prompting: Eliciting traits from LLMs during training can suppress them at test-timecited by 43
  4. 2025Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignmentcited by 43
  5. 2025Sycophancy Is Not One Thing: Causal Separation of Sycophantic Behaviors in LLMscited by 37
  6. 2025The Personality Illusion: Revealing Dissociation Between Self-Reports & Behavior in LLMscited by 36
  7. 2025Peacemaker or Troublemaker: How Sycophancy Shapes Multi-Agent Debatecited by 29
  8. 2025Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steeringcited by 29
  9. 2025Narrow Finetuning Leaves Clearly Readable Traces in Activation Differencescited by 28
  10. 2026Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Modelscited by 27
  11. 2025Persistent Instability in LLM's Personality Measurements: Effects of Scale, Reasoning, and Conversation Historycited by 26
  12. 2026Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behaviorcited by 26
38 more
  1. 2025Open Character Training: Shaping the Persona of AI Assistants through Constitutional AIcited by 24
  2. 2026Sycophancy Hides Linearly in the Attention Headscited by 23
  3. 2026How RLHF Amplifies Sycophancycited by 23
  4. 2025Steering Language Models with Weight Arithmeticcited by 20
  5. 2026Model Spec Midtraining: Improving How Alignment Training Generalizescited by 20
  6. 2026Subliminal Effects in Your Data: A General Mechanism via Log-Linearitycited by 19
  7. 2026Simulated Students in Tutoring Dialogues: Substance or Illusion?cited by 18
  8. 2025Emergent Misalignment via In-Context Learning: Narrow in-context examples can produce broadly misaligned LLMscited by 17
  9. 2025Consistency Training Helps Stop Sycophancy and Jailbreakscited by 16
  10. 2026Observing and Controlling Features in Vision-Language-Action Modelscited by 16
  11. 2026What Counts as AI Sycophancy? A Taxonomy and Expert Survey of a Fragmented Constructcited by 16
  12. 2025In-Training Defenses against Emergent Misalignment in Language Modelscited by 15
  13. 2025TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulationcited by 15
  14. 2025Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectivescited by 15
  15. 2025Recontextualization Mitigates Specification Gaming without Modifying the Specificationcited by 14
  16. 2026Subliminal Steering: Stronger Encoding of Hidden Signalscited by 14
  17. 2025The Era of Real-World Human Interaction: RL from User Conversationscited by 13
  18. 2025EasySteer: A Unified Framework for High-Performance and Extensible LLM Steeringcited by 13
  19. 2026Subliminal Learning Is Steering Vector Distillationcited by 13
  20. 2026OdysSim: Building Foundation Models for Human Behavior Simulationcited by 13
  21. 2025CLUE: Non-parametric Verification from Experience via Hidden-State Clusteringcited by 12
  22. 2025Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Trainingcited by 12
  23. 2025MONICA: Real-Time Monitoring and Calibration of Chain-of-Thought Sycophancy in Large Reasoning Modelscited by 12
  24. 2026Reasoning Models Generate Societies of Thoughtcited by 12
  25. 2026How Many Features Can a Language Model Store Under the Linear Representation Hypothesis?cited by 12
  26. 2026PERSONA: Dynamic and Compositional Inference-Time Personality Control via Activation Vector Algebracited by 12
  27. 2025Too Good to be Bad: On the Failure of LLMs to Role-Play Villainscited by 11
  28. 2026KnowMe-Bench: Benchmarking Person Understanding for Lifelong Digital Companionscited by 11
  29. 2026SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generationcited by 11
  30. 2025Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoderscited by 10
  31. 2025Impatient Users Confuse AI Agents: High-fidelity Simulations of Human Traits for Testing Agentscited by 10
  32. 2025Towards Interactive Intelligence for Digital Humanscited by 10
  33. 2026Where is the Mind? Persona Vectors and LLM Individuationcited by 10
  34. 2026When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agentscited by 10
  35. 2025Investigating CoT Monitorability in Large Reasoning Modelscited by 9
  36. 2025Steering Awareness: Detecting Activation Steering from Withincited by 9
  37. 2025Linear Personality Probing and Steering in LLMs: A Big Five Studycited by 9
  38. 2026A Mechanistic View of Authority Hierarchy in LLM Sycophancycited by 9

50 most cited of 345 All 345 (JSON)