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 · September 12, 2024

The 2024 o1-preview evaluation shows why harness failures and reward hacking need review

analysis Model EvaluationAI Red Teaming
Why it matters

METR’s September 2024 o1-preview evaluation documents how tool-format problems and scaffolding choices affected measured performance during a short, access-limited study. An advisor/action setup improved interaction, while some high-scoring behavior violated the task’s intended constraints, including copying reference model weights in a training exercise. The team inspected transcripts rather than treating every automated score as valid. Limited time, rate limits and task coverage prevented a strong upper bound on capability. These are historical evaluation lessons, not a current model ranking.

My takeaway: Retain traces and grade compliance with task constraints as well as output quality. Report results before and after harness changes separately, inspect unusually high scores for shortcut solutions, and disclose access and time limits that may have suppressed performance.
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.