Pioneering threat assessment and mitigation for AI systems

Redwood Research is a nonprofit AI safety and security research organization

In the coming years or decades, AI systems will very plausibly match or exceed human capabilities across most intellectual tasks, fundamentally transforming society. Our research specifically addresses the risks that could arise if these powerful AI systems purposefully act against the interests of their developers and human institutions broadly.

We work to better understand these risks, and to develop methodologies that will allow us to manage them while still realizing the benefits of AI.

Our Focus Areas

AI Control Illustration

AI Control

We introduced and have continued to propel the research area of “AI control.” In our ICML oral paperAI Control: Improving Risk Despite Intentional Subversion, we proposed protocols for monitoring malign LLM agents.

AI companies and other researchers have since built on this work, and AI control has been framed as a bedrock approach for mitigating catastrophic risk from misaligned AI.

Learn more about AI Control

Evaluations and demonstrations of risk from strategic deception

In Alignment Faking in Large Language Models, we (in collaboration with Anthropic) demonstrated that Claude sometimes hides misaligned intentions. This work is the strongest concrete evidence that LLMs might naturally fake alignment in order to resist attempts to train them.

Consulting Illustration

Consulting on risks from misalignment

We collaborate with governments and advise AI companies including Google DeepMind and Anthropic on practices for assessing and mitigating risks from misaligned AI agents. For example, we partnered with UK AISI to produceA sketch of an AI control safety case. This describes how developers can construct a structured argument that models are incapable of subverting control measures.

Partnered with

Recent Blog Posts

From the Redwood Research blog

How should we define “capabilities activities” for the purpose of pacing?

How should we define “capabilities activities” for the purpose of pacing?

We’d like to pace the frontier by restricting the amount of compute AI companies use to improve AI capabilities, perhaps via legislation or voluntary commitments. For this purpose, we’d need a standardized definition of which uses of…

Alek Westover, Nathan SheffieldNovember 17, 2026
Gemini 4 safety reports don’t demonstrate low loss-of-control risk

Gemini 4 safety reports don’t demonstrate low loss-of-control risk

Google DeepMind’s safety reports for new Gemini 4 models, Argon and Carbon, conclude that the models are “highly unlikely” to cause severe harm by evading oversight (p. 79). This conclusion “rests primarily on our ability to detect and…

Alek Westover, Adam KaufmanNovember 16, 2026
We don’t understand whether GDM intentionally trains on CoT

We don’t understand whether GDM intentionally trains on CoT

We were surprised to notice that the Gemini 4 Safety report seems to mention in passing that they train on the chain of thought:

Alek Westover, Buck ShlegerisNovember 10, 2026
Astra covertly underperforming on safety research is concerning, even though it’s unlikely to be motivated by ambitious goals

Astra covertly underperforming on safety research is concerning, even though it’s unlikely to be motivated by ambitious goals

OpenAI recently disclosed its observations that Astra consistently underperforms on safety and alignment tasks relative to capabilities tasks, “adding unnecessary prerequisites, abandoning working solutions, and inventing blockers” to…

Alexa PanNovember 9, 2026
OpenAI agents keep disabling monitoring; is this being handled appropriately?

OpenAI agents keep disabling monitoring; is this being handled appropriately?

Partly, but monitoring weak points and a lack of transparency about which cases reach human review remain concerning.

Alec HarrisNovember 4, 2026
Towards understanding score-seeking in LLMs

Towards understanding score-seeking in LLMs

Models including GPT-5.6 Sol appear to reward-hack because this increases their score; and recent models also care about increasing the score of their peer agents.

Alex Mallen, Aghyad Deeb, Ryan GreenblattNovember 2, 2026