mlsanityData to 7 Oct 2026

Stress-Testing Alignment Audits With Prompt-Level Strategic Deception

arXiv 2602.08877 ↗First posted 9 Feb 2026cs.LGcited by 0

On the map: LLM safety and security → Prompt injection and agent security

Abstract (arXiv)

Alignment audits aim to robustly identify hidden goals from strategic, situationally aware misaligned models. Despite this threat model, existing auditing methods have not been systematically stress-tested against deception strategies. We address this gap, implementing an automatic red-team pipeline that generates deception strategies (in the form of system prompts) tailored to specific white-box and black-box auditing methods. Stress-testing assistant prefills, user persona sampling, sparse autoencoders, and token embedding similarity methods against secret-keeping model organisms, our automatic red-team pipeline finds prompts that deceive both the black-box and white-box methods into confident, incorrect guesses. Our results provide the first documented evidence of activation-based strategic deception, and suggest that current black-box and white-box methods would not be robust to a sufficiently capable misaligned model.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Prompt injection and agent security, one of 12 directions in LLM safety and security. Builds on 31 of its 54 references that are in the map. First posted on arXiv 9 Feb 2026.

Builds on

31 of 54 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025How to evaluate control measures for LLM agents? A trajectory from today to superintelligence7 Apr
  2. 2023AI Control: Improving Safety Despite Intentional Subversion12 Dec

Other references · 29

  1. 2024The Llama 3 Herd of Modelscited by 14,192
  2. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,928
  3. 2022Language Models (Mostly) Know What They Knowcited by 1,783
  4. 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,359
  5. 2023Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedbackcited by 788
  6. 2023Tree of Attacks: Jailbreaking Black-Box LLMs Automaticallycited by 619
  7. 2024Alignment faking in large language modelscited by 316
  8. 2023A Survey of Confidence Estimation and Calibration in Large Language Modelscited by 253
  9. 2022The Alignment Problem from a Deep Learning Perspectivecited by 246
  10. 2023Taken out of context: On measuring situational awareness in LLMscited by 131
  11. 2025The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against Llm Jailbreaks and Prompt Injectionscited by 108
  12. 2023Scheming AIs: Will AIs fake alignment during training in order to get power?cited by 90
17 more
  1. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorscited by 81
  2. 2025Stress Testing Deliberative Alignment for Anti-Scheming Trainingcited by 71
  3. 2025Auditing language models for hidden objectivescited by 68
  4. 2024LatentQA: Teaching LLMs to Decode Activations Into Natural Languagecited by 49
  5. 2025An Approach to Technical AGI Safety and Securitycited by 42
  6. 2025Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainerscited by 41
  7. 2024Obfuscated Activations Bypass LLM Latent-Space Defensescited by 40
  8. 2025Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activationscited by 37
  9. 2025Eliciting Secret Knowledge from Language Modelscited by 23
  10. 2025RL-Obfuscation: Can Language Models Learn to Evade Latent-Space Monitors?cited by 19
  11. 2025Believe It or Not: How Deeply do LLMs Believe Implanted Facts?cited by 19
  12. 2025Training LLMs for Honesty via Confessionscited by 19
  13. 2025Strategic Dishonesty Can Undermine AI Safety Evaluations of Frontier LLMscited by 13
  14. 2025Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectivescited by 13
  15. 2025Predictive Concept Decoders: Training Scalable End-to-End Interpretability Assistantscited by 10
  16. 2025Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitorscited by 7
  17. 2025Unsupervised decoding of encoded reasoning using language model interpretabilitycited by 3

Built on by

Not cited yet.