European Commission Beefs Up AI Office Ahead of AI Act Enforcement Deadline

The European Commission announced Friday the creation of an expanded team to enforce the EU's artificial intelligence rules. Thirty-eight new staff will join the AI Office in Brussels just before the AI Act's rules on general-purpose model providers come into full effect on Sunday.

What Changed

Until now, the AI Office mainly served an advisory and coordinating role. Starting August 2, it gains real investigative tools: it can demand documentation from model providers, conduct interviews with company staff, commission independent assessments of systems and, as a last resort, impose financial penalties or restrict access to the EU market.

The staffing boost is meant to make real use of these powers possible. The existing team overseeing general-purpose models was, according to the Commission itself, too small for the scale of its responsibilities - the AI Office's safety division alone had employed around 40 people out of more than 140 staff in total.

New Powers

Article 101 of the AI Act gives the Commission four separate avenues for imposing penalties on general-purpose model providers: for breaching GPAI substantive rules, for ignoring requests for documentation, for refusing to make a model available for assessment, and for failing to implement ordered corrective measures. The maximum fine reaches 3 percent of a company's global annual turnover or 15 million euros, whichever is higher.

The new team will focus on specific categories of violations: publishing AI-generated sexual content, fake photos and videos, and cyberthreats targeting public infrastructure. So-called systemic risks will also come under scrutiny, including scenarios involving chemical, biological, radiological and nuclear threats.

As enforcement begins, we are taking an important step toward artificial intelligence that people and businesses can understand and trust, and whose benefits are widely shared across our society - Henna Virkkunen, Vice-President of the European Commission for Technological Sovereignty

Who's Under Scrutiny

The team will oversee both small startups and the largest model providers operating in the European Union - among those named are OpenAI, Anthropic and the Chinese company DeepSeek, while Amazon, Google and Microsoft also come up in the context of the market's broader dependence on large platforms. Companies will be required to clearly inform users, through labels or digital watermarks, when content - text, image or audio - was created using artificial intelligence.

Alongside traditional inspections, the Commission is launching two new reporting channels: a Whistleblower Tool for tech industry employees who want to confidentially report illegal practices by their employer, and a Compliance Tool for ordinary users flagging violations. This is meant to close the information gap that NGOs monitoring the AI Act's rollout have complained about for months.

What This Means for Businesses in Poland

For Polish companies deploying solutions built on foreign providers' models, this poses a real business risk from the other end of the chain - if a model provider (e.g. GPT, Claude or Gemini) becomes subject to proceedings or is forced to restrict features, it could affect the availability and terms of use of tools that companies have already built their processes around. The strengthened oversight also increases pressure for accurate labeling of AI-generated content, which applies to any company publishing marketing or customer service materials created with the help of language models.

Poland's law implementing the AI Act has been signed by the president, and the national commission overseeing artificial intelligence is set to launch in November. The expansion of the EU's AI Office does not replace national oversight, but it sets a standard the Polish regulator will also need to align with, especially when it comes to cooperating on cross-border proceedings against global model providers.

What's Next

The AI Office's full powers take effect alongside the requirement to label AI-generated content, which will become mandatory across the European Union on August 2, 2026. In the coming weeks, the first formal requests for documentation to the largest model providers can be expected, and eventually the first investigations and possible penalties, which will show whether the new, strengthened team can actually enforce the rules against companies operating globally, outside the EU's direct jurisdiction.