Saint-Josse-ten-Noode: The European Commission’s top scientific and ethics advisers called on 16 November 2026 for Europe to build its own independent AI evaluation capacity, warning that traditional guardrails no longer suffice for the next generation of artificial intelligence. The Group of Chief Scientific Advisors and the European Group on Ethics in Science and New Technologies issued the call in a rare joint statement.
The paper, titled “The next phase of artificial intelligence – Governing for safety, security and human agency”, argues that Europe need not choose between innovation and safety. Instead, the advisers say the EU should treat safety, security and human agency as prerequisites for developing AI, not as obstacles to it.
Their central recommendation on independent AI evaluation targets who checks the models. The advisers want the EU to develop independent AI evaluation in-house and to stop delegating the testing of powerful systems to the commercial developers that build them. Evaluation methods, they add, must evolve as fast as AI capabilities do, which requires a permanent ecosystem for safety, security and continuous learning.
The statement sets out four lines of action. Lawmakers should keep the rules strong, calibrate the pace of development and deployment to the level of potential danger, build public capacity for independent AI evaluation, and fund that wider safety ecosystem. The advisers ground their case in the AI Act, Regulation (EU) 2024/1689, the 2025 General-Purpose AI Code of Practice and the Commission’s proposal for a Cloud and AI Development Act, COM(2026) 502.
The timing gives the warning political weight. Several capitals and industry groups have pushed the Commission to simplify digital rules and to slow parts of the AI Act’s rollout. The advisers caution that weak political commitment or thin institutional capacity could hollow out existing obligations, and that efforts to delay or weaken legislation could compromise safeguards.
The seven chief scientific advisers serve in a personal capacity and normally answer specific requests from the College of Commissioners. A joint text with the ethics group is unusual, and the Commission’s research directorate presented it as a deliberate signal. The two groups previously advised on AI in science, AI in crisis management and the governance of neuro-AI.
For research policy, independent AI evaluation implies public investment in testing infrastructure, from compute access for evaluators to trained staff in the EU AI Office and national authorities. It also raises questions for Horizon Europe and the next framework programme about how much money should flow to safety science rather than to model development alone.
The advisers did not set a deadline, and the Commission has not said how it will respond. Still, the joint statement puts independent AI evaluation at the centre of the debate on Europe’s digital rulebook, and any future simplification package will now have to show that it keeps public testing capacity intact.





