A letter from the staff committee of the Court of Justice of the European Union demands the EU Commission explain its plans to use AI to deal with an increase in candidates for EU jobs. The committee deplores the fact that plans for the AI tool, built on tech developed by US AI company Anthropic and hosted on Amazon Web Services (AWS) cloud infrastructure, have been kept hidden, leaving the committee unable to raise concerns sooner.
A record number of people, over 170 000, applied this year for the entrance exams to become permanent administrators in European Union institutions. With only 1490 open vacancies, the selection will take considerable time and resources. For the first time, selection won’t be supervised solely by the European Personnel Selection Office (EPSO) and the individual institutions. Instead, under a pilot initiative, the demanding process will be complemented with an AI tool to screen candidates.
An article published by Euractiv revealed on 15 June the plan to use an AI tool called Job Matching Application, developed by the private company Accenture with Anthropic’s large language model (LLM) and hosted on Amazon Web Services (AWS) cloud infrastructure. The tool is scheduled to be tested this summer on the first pool of candidates. The staff Committee of the EU Court claims to have only been made aware of this project through the press. Now, the committee is raising concerns in its 2 July letter regarding compliance with the principles of equal treatment, non-discrimination, transparency and good administration. It also addresses the high level of personal data protection that should be implemented before deployment.
High-risk tool and uncertainty over bias
The Commission launched the pilot project the same year it adopted its landmark rules on AI. The EU claims these rules represent the world’s first comprehensive legal framework for AI. It should, then, have set the bar for the EU to lead in transparency and sensible use. The letter argues, however, that the EU has failed at the first post. It points out that that the Commission: “whilst working to develop a legal framework designed to protect digital sovereignty and fundamental rights, has simultaneously launched a project that jeopardises them”.
The AI Act classifies AI tools deployed in recruitment as high-risk because they can reproduce and amplify bias and therefore require close human control, transparency and risk management. The letter warns that even if the final decision is made by a human, selection and recommendations made in advance by technology can be “decisive”. It is therefore essential to know and monitor the criteria that the AI uses to screen candidates. EU institutions have a duty to adhere strictly to fairness and transparency in their selection processes, to guarantee the principles of equality and non-discrimination for candidates.
The staff committee also raises the matter of the reliability of the tool deployed. An investigation of the civil society organisation Privacy International published on 9th July, for example, revealed that two AI powered recruitment platforms it investigated relied on “opaque and often unreliable systems.”
A threat to EU digital self-determination
In June last year, the European Commission unveiled its EU Tech Sovereignty Package to reduce dependencies on foreign tech and maintain control over critical technology and infrastructure. One would think that the process of selecting who works in the EU’s institutions would be a sensitive role and that giving it to US companies could therefore present a serious liability. This is the opinion of the staff committee of the EU Court, who say in the letter:
“Recruitment (…) plays a key strategic role, as it serves to select the individuals who will be called upon to participate, directly or indirectly, in the functioning of the Union’s institutions and bodies.”
The letter goes on to question control over data and, therefore, the autonomy of EU institutions. The safeguard given by the EU Commission in the press about the program that candidates’ personal data will not be used to train AI models does not satisfy the members of the court and the authors of the letter. They flatly denounce the Commission’s choice to give responsibility for “high-risk processing operations to select future civil servants and staff” to “companies that are already threatening the sovereignty of the EU”.
The letter raises another issue regarding the EU Court’s own duty of impartiality. Considering that it may need to examine cases on AI deployment in recruitment, its own participation in deploying such a tool to hire its permanent staff could pose a serious conflict.
The letter raises serious questions about this pilot project. In particular, it demands greater scrutiny on the measures taken by the EU Commission to make sure the project respects EU data protection, fundamental rights and the EU AI Act.
