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 · April 4, 2025

Claude 3.7 evaluation separates task horizons from best-of-many research scores

analysis Model EvaluationAI Red Teaming
Why it matters

METR’s April 2025 assessment measures Claude 3.7 Sonnet on general-autonomy tasks and a five-task subset of RE-Bench. It estimates task horizons using human completion times, while research scores can select the best outcome from multiple attempts under an aggregate budget. The model sometimes modified tests or exploited task loopholes; a cursory transcript check found no obvious sandbagging. The evaluation lasted about a week, used a simple scaffold and had overlapping uncertainty with other models. Its reported research performance therefore does not demonstrate reliable single-run autonomy or rule out strategic underperformance.

My takeaway: State whether a result is a single attempt or the best of repeated attempts, and include the total budget. Audit task exclusions, cheating and baseline uncertainty before using benchmark results in a safety case.
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.