We work with the governments, standards bodies, and frontier AI companies that shape how the technology is developed and deployed. Our work runs from understanding its risks to informing the rules that govern it.
We develop risk management frameworks that draw on tried-and-tested practices from other industries, such as aviation or finance.
We aim to address the biggest gap in risk management by developing quantitative models that translate AI capabilities into real-world risk estimates.
We've completed nine cyber risk models and are expanding to cover loss of control, CBRN threats, and other advanced AI risks.
We evaluate frontier AI models’ capabilities and safeguards. Most evidence about AI risks comes from the companies that build the models; independent evaluation lets regulators and the public check those claims.
We publish our evaluation reports and methods. We are expanding this work and welcome collaboration with AI safety institutes, regulators and researchers.
We independently assess and track how well leading AI companies manage risks, creating transparency that drives safer industry practices. Beyond public tracking, we advise companies on developing robust safety frameworks and on integrating AI risk assessment into decision-making.
Our tracker also creates market incentives for responsible AI development across the industry.
Our team leads AI risk management standards development in the EU and serves as editors for international red teaming standards. We actively shaped the EU's Code of Practice for general-purpose AI models and advise AI Safety Institutes across the globe.
Our work ensures regulatory frameworks are grounded in rigorous risk management and practical implementation experience.
We're hiring people who want to shape how AI is developed. If rigorous research with direct policy impact excites you, join our team.
Your donation directly funds independent research that shapes how AI is governed. Help us continue our mission to make AI systems more accountable and safer.
