The European Commission's AI Office published a call on 12 June 2026 for experts to join an online workshop, tentatively set for 15 July. The workshop concerns independence and qualification requirements for external evaluators of general-purpose AI (GPAI) models with systemic risk. The requirements would apply to outside organizations that evaluate models and assess systemic risks under the GPAI Code of Practice. The AI Office says the workshop's input will inform requirements meant to ensure the credibility and reliability of external evaluations under the AI Act.[1]
Applicants had until 21 June to express interest, and invitations were due by 7 July. The workshop is not meant to cover the requirements the AI Act sets for its scientific panel or for independent experts the Commission may appoint to conduct evaluations under Article 92(3).[1]
The Commission describes the Code as a voluntary tool that providers can sign to show they comply with the AI Act.[2] Its Safety and Security chapter already commits signatories to have adequately qualified independent external evaluators assess their models' systemic risks. A signatory may go without them if its model counts as similarly safe or safer under the Code's criteria. It may also do so if it fails to appoint suitable evaluators despite early search efforts, such as a public call open for 20 business days. In that case, it must account for the added uncertainty when judging whether the risks are acceptable. Qualification requires significant domain expertise and evaluation experience, appropriate information security protocols and an agreement to protect confidential information. The Code treats an evaluator as independent if it has no financial, operational or management dependence on the signatory and is otherwise free from the signatory's control in reaching conclusions. Signatories explain their choice of evaluator against the qualification criteria in their model reports to the AI Office.[3]
Sources
Check or correct this page
Verification record
The digest below shows whether the text you are reading matches the published source. It does not show that the reporting is correct. The build computes it from this page's Markdown source and lists it in the site manifest.
0e420bfe8b800a879c9e65caab6679abb74920d1aca23038495ec24cd7eb14b1
.jpg){kind=link}