Commission seeks experts for AI Scientific Panel

Context

This panel is being set up by the European Commission to advise and support the European AI Office and national authorities, particularly concerning the implementation and enforcement of the 'AI Act' (Regulation (EU) 2024/1689).

Areas

Here is a concise summary of the key areas outlined:

Evaluation of GPAI Capabilities and Impacts: Assessing model performance and risks through testing, impact studies, rights assessments, economic analyses, and forecasting.
AI Risk Assessment Methodologies: Developing frameworks to identify, analyse, model, estimate, and evaluate AI risks, including taxonomies and safety cases.
Technical Risk Mitigations and Best Practices: Implementing safety measures like fine-tuning, filters, watermarking, data handling, governance, incident response, and audits.
Systemic Misuse and Deployment Risks: Addressing GPAI misuse across high-risk domains (e.g., CBRN, manipulation, fairness, public safety).
Cyber Offense Risks: Understanding how GPAI can be exploited for cyberattacks, including zero-days, social engineering, and infrastructure targeting.
GPAI Provider Security: Securing models against leaks, theft, unauthorised releases, and circumvention of safeguards.
Emergent Systemic Risks: Investigating new risks from model behaviors, and capabilities like misalignment, deception, and self-improvement.
Compute Measurement and Thresholds: Developing standardised methods to measure, report, and verify compute usage in model training.

Target Groups

Independent experts selected in a personal capacity; these individuals must be highly qualified, multidisciplinary, and interdisciplinary experts with a sufficient level of up-to-date scientific, socio-technical, or technical expertise on various aspects related to AI, its impacts, or otherwise relevant to the effective enforcement of the AI Act.

Eligibility Criteria

Multidisciplinary and interdisciplinary adequate and up-to-date scientific, technical, or sociotechnical expertise related to AI systems or general-purpose AI models, their impacts, or aspects relevant to the enforcement of Regulation (EU) 2024/1689.
Independence from any provider of AI systems or general-purpose AI models: Candidates must demonstrate they are not employees of, or in a contractual relationship with, such providers at the time of expressing interest and throughout their term of office.
Impartiality and objectivity.
An ability to carry out activities diligently, accurately, and objectively.

Specific requirements for eligibility include:

A PhD in a relevant area or equivalent experience.
Proven scientific impact on AI/GPAI research or the study of AI impacts.
At least 4/5ths of the experts should be nationals from EU Member States or European Free Trade Association (EFTA) members that are also members of the European Economic Area. However, EU citizenship is not a strict requirement, as nationals of non-EU countries can be appointed. The panel aims to include at least one and no more than three nationals from each EU Member State and EFTA/EEA member.
The Commission aims to ensure fair gender balance in the selection of experts.
All candidates must submit a declaration of interests (DOI) to demonstrate independence and disclose any potential conflicts. Selected members must also sign a declaration of confidentiality. The DOI of selected members will be publicly available.