AI Standards Lab is an independent non-profit dedicated to accelerating the development of AI safety standards.
We are structured as a virtual lab, bringing together experts from all over the world. Our team has expertise in computer science, AI research, safety engineering, standards development, and law.
EU AI Act related activities
We contribute to the writing of many standards-type documents in the context of the EU AI Act. These documents give more technical details about how organisations can achieve the safety outcomes required by the Act. The documents we have worked on and are working on include:
- The EU’s General-Purpose AI Code of Practice,
- The technical safety standards under development in the CEN-CENELEC JTC21 committee, and
- Guideline documents on various topics written by the European Commission and the EU AI Office.
Our main focus is to ensure that all these documents define clear minimum safety requirements. Clear minimum requirements are crucial in preventing a race to the bottom in safety, and to enable consistent enforcement across the market.
Our work in support of the EU AI Act spans a broad range of risks. We work on society-level existential and systemic risks, risks to physical and mental health, risks to human rights, and risks to democracy and the rule of law.
Other activities
Beyond work specific to the EU, we run projects aimed at advancing the frontier of AI safety engineering globally. We focus on areas where we believe progress can be accelerated through the broad and voluntary adoption of more standardised approaches. These include:
- Defining standardised approaches for quality assessment of frontier model evaluations, and
- Defining best practices for analysing and mitigating loss of control risks for frontier models
We also publish reports and papers, often in collaboration with other authors and organisations. This work aims to inform AI safety engineers, AI policymakers, and parties seeking to understand the role of standards in AI safety engineering.
History
AI Standards Lab was founded in November 2023 and incorporated as a US non-profit in April 2024. Our work is funded by charitable grants, which pay for a team size of approximately five full time equivalents. The Lab’s mission and working methods were developed with a volunteer team, in an AI Safety Camp project running from March to June 2023.
See our public outputs page for an overview of our work. Our contributions to standards work in the CEN-CENELEC JTC21 committee are not included because they are confidential under committee rules.