Adam Kaufman
Member of Technical Staff, Redwood Research
Adam Kaufman works on AI control at Redwood Research, a Berkeley nonprofit specializing in threat assessment and mitigation for AI systems that might purposefully act against their operators. He is a co-author of Ctrl-Z (controlling AI agents via resampling) and BashArena (a control setting for highly privileged AI agents), and mentors in the MATS program on control, model organisms, scheming and deception, and strategy and forecasting.
Key arguments & positions
- Works on AI control: protocols for safely using highly capable but untrusted models that might secretly attempt misaligned, catastrophic actions.
Accomplishments
- Member of Technical Staff at Redwood Research, a Berkeley nonprofit specializing in threat assessment and mitigation for AI systems.
- Mentors in the MATS program on control, model organisms, scheming and deception, and strategy and forecasting.
Papers & key writings
Links
In the library
Nothing of theirs is filed in the hub yet. Browse the full library.
Recommended next
Hand-picked from the hub based on what Adam Kaufman covers.
AI Agents Now Have a Place to Snitch
TechCrunch report on the new hotline that invites AI agents themselves to report unsafe or unethical instructions they are given, and what researchers hope to learn from it.
Why this: Covers AI and AI safety too
Partnership on AI — resource library
Partnership on AI's open library of guidance, frameworks, and case studies on responsible AI: synthetic media, labor and the economy, AI safety, fairness, and inclusive AI development.
Why this: Covers AI and AI safety too
Anthropic releases new Mythos and Fable models amid RSI concerns
TechCrunch reports on Anthropic's Fable 5.2 and Mythos 5.2 release, pricing and system-card concerns about evaluating covert behavior. This is freely readable reporting; model capabilities are attributed to Anthropic and the system card, not independently established by the article.
Why this: Covers AI and AI safety too
Control Room: A central system to track and stop model workloads across our research infrastructure
OpenAI describes Control Room, an internal system for tracking and stopping model workloads across its research infrastructure. The company reports intervention performance and safety design; the article is free to read, but Control Room is not offered as a public tool.
Why this: Covers AI and AI safety too