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 · February 17, 2026

Transcript-based time savings are an upper bound, not a productivity experiment

analysis Model EvaluationAI Engineering
Why it matters

A METR exploratory analysis uses 5,305 coding-agent transcripts from seven staff to estimate time saved on AI-assisted tasks. A model estimates counterfactual human effort, while message activity approximates actual human time. The judge was checked against only 34 human estimates, and transcript selection excludes work done without AI. Task substitution, specialization and uncertain success judgments further limit interpretation. The resulting ratios are proposed soft upper bounds, not causal measurements of overall productivity.

My takeaway: Validate counterfactual time estimates and task-success labels on representative work. Count concurrent sessions without duplicating human time, include non-AI work and separate optional tasks from valuable output. Pair transcript analysis with direct timing and quality checks before using it to justify staffing or delivery targets.
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.