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 · November 22, 2024

RE-Bench compares research agents and human experts under explicit compute budgets

analysis Model EvaluationAI Engineering
Why it matters

METR’s November 2024 RE-Bench release provides seven research-engineering environments, human-expert attempts and agent transcripts. Tasks involve objectives such as optimizing kernels, recovering model performance and fitting scaling laws under specified resources. Results depend strongly on time allocation: agents often benefit from selecting the best of many short attempts, while humans benefit from longer continuous work. The environments offer clearer goals and faster feedback than much real research, and cheating requires inspection. The benchmark measures bounded engineering performance, not autonomous completion of an open-ended research program.

My takeaway: Report continuous-run performance alongside best-of-many results and total compute budgets. Preserve independent scoring, inspect rule violations and test longer feedback loops before extrapolating benchmark gains to research automation.
Keep exploring

More curated notes connected through Model Evaluation and AI Engineering.

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.

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.

OpenAI News · framework

OpenAI’s Frontier Governance Framework

OpenAI's 22-page Frontier Governance Framework maps its frontier-model processes to California's Transparency in Frontier AI Act and the EU AI Act's general-purpose AI code. It documents lifecycle risk assessment, cyber-offense and other risk tiers, mitigation and residual-risk decisions, critical-incident handling, security risk management, model reporting, external review, responsibility allocation, and change control.