EU adopts the Scientific Panel Implementing Act under the AI Act

On 11 March 2025, the European Commission adopted the implementing act that establishes a scientific panel of independent experts under the EU AI Act. The act lays down the rules and procedures for establishing and operating the panel, which will support enforcement of the AI Act.

What it is

The scientific panel is an expert advisory body established by the European Commission in accordance with Article 68 of the AI Act, in conjunction with Commission Implementing Regulation (EU) 2025/454. It is meant to play a key role in AI Act governance, offering technical advice and input on enforcement, including the ability to alert the AI Office of risks posed by a general-purpose AI model. Sixty world-class independent AI experts have since been recruited to join the panel.

What it requires

The panel's core responsibilities include alerting the AI Office to systemic risks, advising on GPAI classification and evaluation methodologies, and supporting market surveillance activities. Its Article 68 mandate includes alerting the AI Office of possible systemic risks at Union level of general-purpose AI models, in accordance with Article 90, and contributing to the development of tools and methodologies for evaluating capabilities of general-purpose AI models and systems, including through benchmarks. Experts are appointed for a 24-month term, renewable, in a personal capacity, and each must draw up a declaration of interests that is made publicly available. Candidates were expected to have expertise in GPAI models and systems, AI impacts, or related fields, such as model evaluation, risk assessment and mitigations, cybersecurity, emerging systemic risks, and compute measurements and thresholds, and to hold a PhD or equivalent experience while remaining independent of any AI provider. The selection process also weighed geographic balance, limiting membership to at most three nationals from each country, and gender balance, giving preference to the underrepresented gender when candidates were equally qualified.

What security leaders should do

Security and compliance teams at providers of general-purpose AI models should track the panel's alerting role closely, since a qualified alert to the AI Office of possible systemic risks at Union level can precede enforcement attention on a model. Teams building model evaluation and red-teaming programs should anticipate that panel-developed benchmarks for evaluating GPAI capabilities may inform how regulators expect systemic risk to be measured. Because the panel also advises on classification of GPAI models with systemic risk, security leaders should keep documentation current enough to respond to classification questions that could change which AI Act obligations apply to their model.