Research institute · Research orgs · Berkeley, California · Est. 2021
Redwood Research is a registered 501(c)(3) nonprofit AI safety and security research organization focused on threat assessment and mitigation for powerful AI systems that could purposefully act against the interests of their developers and human institutions. Redwood’s technical work centers on “AI control”—research into control protocols and evaluation methods designed to keep systems safe even when models are intentionally misaligned or attempting to subvert oversight.
The organization also conducts and publishes research on strategic deception and alignment-faking risks, including an empirical collaboration with Anthropic, and it runs applied safety work such as consulting with governments and advising AI companies on risk assessment and mitigations for misaligned agents. In addition, Redwood emphasizes empirical and adversarial approaches to measuring monitorability and control robustness, including through its control evaluations and research artifacts (e.g., BashArena and a public Control Leaderboard).
No curated programs yet. Check the careers page ↗.
Publications are not yet indexed for this organization. Read them on redwoodresearch.org ↗.
Redwood reports measurements suggesting GPT-6 Astra can perform more substantial cognition without verbalizing it in chain-of-thought when prompts include filler tokens, raising monitoring concerns.
CoT controllability evals seem very under-elicitedRedwood argues that chain-of-thought controllability evaluations may be under-elicited and that prompt optimizations can materially change measured performance.
An operationalization of opaque serial depthRedwood describes an operationalization (a proxy metric) for “opaque serial depth” between text bottlenecks as a way to reason about latent-reasoning-like architectures and monitorability risks.
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident (summary post)A shorter Redwood blog post pointing readers to the full report and summarizing key takeaways.
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incidentRedwood publishes its independent investigation report with a detailed timeline and scope, including datasets reviewed and methods for analyzing agent collaboration and behavior during the stated in-scope window.