Redwood Research blog: Who It's For and Who Sponsors It

Covers: AI alignment research, monitoring systems, red-teaming techniques, evaluations, agent oversight, and mitigation of catastrophic risks.
For: AI safety researchers, machine learning engineers, and technical leaders building frontier AI models.
Best for: AI developers and researchers designing safety evaluations and monitoring tools for advanced models.
Skip if: General business professionals seeking basic generative AI adoption tips.

Start with

Simulated Control Evaluations: Explains methods for testing whether AI monitors catch agents attempting to cause harm.
Red-Teaming OpenAI's Internal Agent Monitoring Systems: Shares findings from a high-access stress test of frontier model internal monitoring systems.
Screwtape: Adversarial Transcript Editing for Monitor Red-Teaming: Demonstrates a pipeline that generates realistic attacks to test and evade AI monitors.

Summarised by AI from 12 recent issues. Audience size is not included. Prices are not included. Report a mistake.