Digital Sovereignty

The Commission Hired US AI to Vet Its Own Staff

Brussels used Anthropic's Claude on AWS to screen 170,000 EU job applicants - and only reviewed European alternatives after staff objected.

Digital SovereigntyBy Servola Tech Desk2026-08-074 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

The Commission Hired US AI to Vet Its Own Staff

Key takeaways

The European Commission used Accenture's Job Matching Application, built on Anthropic's Claude and hosted on AWS, to screen more than 170,000 applicants for 1,490 EU administrator posts - a US technology stack operating inside the institution that writes Europe's digital sovereignty rules.
EU staff first learned of the project from Euractiv's reporting on 15 June 2026, not from any internal announcement by the Commission.
On 2 July 2026 the Court of Justice of the EU staff committee formally objected, warning the tool 'jeopardises' the digital sovereignty and fundamental rights the Commission is meanwhile trying to legislate.
A Commission spokesperson confirmed on 3 August 2026 that it is now 'evaluating' replacing the US-based infrastructure with European alternatives - a review that began after the leak, not before.
Employment screening sits in the high-risk category under the EU AI Act's own risk framework, which makes the Commission's choice of vendor a compliance question as much as a political one.

A Hiring Tool Built the American Way

The European Commission needed to screen candidates for permanent administrator positions in its own civil service, so it built a tool to do it. The Job Matching Application was developed by Accenture, the private contractor, running on Anthropic's Claude as the underlying AI model and hosted on Amazon Web Services. Testing is planned for summer 2026.

The scale is not small. The system is meant to screen more than 170,000 applicants competing for 1,490 open vacancies. That is the kind of volume no human recruitment team processes by hand, which is exactly why the Commission reached for an AI system in the first place.

Staff Found Out From the News, Not From Brussels

EU staff did not learn about the Job Matching Application through any internal briefing, works council notice, or staff bulletin. They learned about it on 15 June 2026, when Euractiv reported the story.

For an institution that positions itself as a careful steward of employee data and algorithmic transparency, the sequence matters as much as the substance. The people whose applications the tool would screen found out the same way the general public did.

The Court's Staff Committee Pushes Back

On 2 July 2026 the staff committee of the Court of Justice of the EU sent a formal letter of objection. Its central line was blunt: the Commission, the committee wrote, 'whilst working to develop a legal framework designed to protect digital sovereignty and fundamental rights, has simultaneously launched a project that jeopardises them.'

The letter raised four distinct concerns: a lack of transparency around how the tool was built and deployed, the risk of bias in an AI-driven screening process, data-protection questions, and the broader threat to EU digital sovereignty from relying on US companies for something this sensitive. It also noted that although a human makes the final hiring call, AI-generated recommendations can be 'decisive' in practice, which puts the reliability of an opaque algorithm at the center of a real employment decision.

Brussels Says It Is Looking at European Options

The Commission's response came on 3 August 2026, after Euractiv's reporting had already forced the issue into the open. A spokesperson confirmed to Euractiv that the executive is 'evaluating the possibility of replacing the US-based digital infrastructure with European alternatives,' and gave assurance that candidate data would not be used to train AI models.

EU Executive Vice-President Henna Virkkunen told the Financial Times that 'artificial intelligence has become a geopolitical instrument, making it essential for Europe to develop alternatives to US AI models.' The statement lands one month after Brussels announced measures in June 2026 to strengthen what it called strategic sovereignty by reducing dependence on American technology across semiconductors, cloud computing, and AI.

The Real Lesson: Sovereignty Is a Choice, Not a Default

Why it matters: the Commission is the author of the EU AI Act and the architect of Europe's digital sovereignty policy. When it needed to build something real, for its own staff, on a deadline, it reached for Anthropic and AWS rather than a European alternative. That is not hypocrisy so much as a demonstration of how procurement actually works under time pressure: capability wins.

The review of European alternatives only started once the project became public and the Commission's own staff objected. It was not triggered by a compliance requirement, a mandatory sovereignty audit, or a proactive risk assessment. It was triggered by a press leak and an internal letter.

Employment screening is treated as a high-risk use case under the EU AI Act's own risk framework, which means the Commission's vendor choice was never a purely technical decision. For any EU or UK business facing the same buy-sovereign-or-buy-capable tradeoff in HR, finance, or customer data systems, this is the honest state of play in mid-2026: sovereignty compliance is a decision you make on purpose, not one the default vendor list makes for you.

Frequently asked questions

What is the Job Matching Application?

It is an AI-based screening tool the European Commission built to process applications for permanent administrator positions in the EU civil service. Accenture built it, Anthropic's Claude is the underlying AI model, and it runs on Amazon Web Services. It was set to screen over 170,000 applicants for 1,490 vacancies, with testing planned for summer 2026.

Why did the CJEU staff committee object?

In a letter dated 2 July 2026, the committee argued the Commission was undermining the very digital sovereignty and fundamental-rights framework it is drafting. It cited a lack of transparency, bias risk, data-protection concerns, and dependence on US companies, and warned that AI-generated recommendations can be decisive in practice even though a human signs off on the final decision.

Is the Commission replacing the US infrastructure?

It has not committed to doing so. A Commission spokesperson said on 3 August 2026 that the executive is evaluating the possibility of replacing the US-based digital infrastructure with European alternatives, and confirmed candidate data would not be used to train AI models. No decision or timeline has been announced.

The Commission did not set out to undermine its own sovereignty policy. It set out to hire people, on a deadline, at scale, and it reached for the tools that worked. That is precisely the point: sovereignty loses to capability by default, inside Brussels as much as anywhere else, and the only way to change that outcome is to decide it in advance, before the next staff committee has to write the letter.

European CommissionAnthropicAWSEU AI ActDigital SovereigntyAccenture

1,134 Lab Staff Say The Brake Does Not Exist

Chief scientists at OpenAI, Anthropic, Meta and Google signed one statement asking Washington to build pacing tools. Their argument is that no such tool exists.

6 min read

Digital Sovereignty

A Chinese Model May Be Answering Your Prompts

On OpenRouter, Chinese-built models have taken over 30 percent of US company tokens since February, peaking near 46 percent. If you route through an aggregator, you cannot say which model read your data.

3 min read

Decisions

The Codename Worked. The Trading Controls Did Not

An indictment unsealed on 24 July says two Volkswagen engineers traded on a joint venture known internally as Project Climb. The control that failed is a common one.

5 min read

← All articles