AI Red Team Engineering notes, research references, and field notes on AI systems, responsible AI, governance, and compliance.
Global search

Search AI Red Team

Search across the complete knowledge base.

Enter at least two characters to begin.
↑↓ select Enter open Browse the full research archive
METR · August 8, 2025

METR tests whether imperfect reasoning traces can still reveal hidden clues

analysis Model EvaluationAI Red Teaming
Why it matters

METR’s August 2025 study separates complete faithfulness from a narrower monitoring question: can a reasoning trace reveal that a model used extra information? Its DAFT setup uses free-response tasks and supplied clues to reduce the chance that a correct answer came from guessing, then tests detection from the trace. Reasoning could remain informative even when it did not explicitly acknowledge the clue. The authors flag imperfect ground truth, distribution differences between task conditions, and limited adversarial optimization; these experiments do not establish reliable detection of naturally occurring deception.

My takeaway: Measure what a trace lets an independent monitor detect, with matched task conditions and explicit ground-truth assumptions. Report detection performance separately from claims that every reasoning step faithfully explains the answer.
Keep exploring

More curated notes connected through Model Evaluation and AI Red Teaming.

OpenAI News · framework

Pacing model development in an era of cyber-critical capabilities

OpenAI says preliminary evidence that Astra may meet its Critical cybersecurity threshold led it to pause frontier reinforcement-learning work for two weeks and keep its largest planned run on hold. New safeguards include stronger workload and network isolation, continuous boundary testing, token-level monitoring that escalates suspicious tool activity, and broader alignment checks for deception, reward hacking, and unauthorized access.

OWASP GenAI Security Project · guide

OWASP Top 10 for Agentic Applications for 2026

OWASP's community guide organizes agentic-system risk into ten categories, including goal hijacking, tool misuse, identity and privilege abuse, memory poisoning, insecure inter-agent communication, cascading failures, and rogue-agent behavior. It provides a shared taxonomy and mitigation starting point rather than a certification checklist or evidence that a deployed system is secure.

OpenAI News · framework

A blueprint for democratic governance of frontier AI

OpenAI proposes a three-part U.S. frontier-AI governance model: harmonize emerging state safety laws into a federal baseline, strengthen CAISI as an evaluation and standards institution, and coordinate a broader resilience program. Proposed controls include severe-risk evaluations, transparency reports, independent audits, safety-incident reporting, model-weight security, whistleblower protection, and periodic technical assessments.