Skip to content

Europe · Analysis

Independent · Brussels & Berlin

Technology · Digital regulation

EU tech chief signals possible delay to parts of AI Act as guidance falls behind

Henna Virkkunen tells digital ministers that August deadlines for general-purpose AI models could slip if codes of practice and technical standards are not finalised in time.

By , Technology Editor

Published

7 min read

The European Commission's most senior technology official has publicly acknowledged that the EU's landmark artificial intelligence legislation may not meet its own timetable. Speaking to digital ministers gathered in Luxembourg on Friday, Executive Vice President Henna Virkkunen said the bloc should not rule out postponing elements of the AI Act if the standards and guidance companies need to comply are not finalised in time.

Commission opens door to AI Act delays

The admission marks a notable shift in tone. Since the regulation was agreed in December 2023 after marathon negotiations between the European Parliament and the Council, the Commission has consistently emphasised that the phased implementation schedule, first prohibitions in February 2025, general-purpose AI model obligations in August 2025, and the full regime in August 2026, was fixed and credible. Virkkunen's comments, made during an informal meeting of ministers under the Polish Council presidency, suggest that certainty is fraying at the edges.

The immediate pressure point is the 2 August 2025 deadline for providers of general-purpose AI models. These companies, among them the largest US and Chinese developers as well as European firms, must comply with transparency, copyright and systemic-risk obligations set out in Chapter V of the regulation. Yet the code of practice intended to translate those obligations into actionable steps remains in draft form, and the harmonised technical standards that would give companies a presumption of conformity have not been published.

Industry pressure and the stop-the-clock demand

Industry associations have spent months arguing that compliance is impossible without finished guidance. In recent weeks they have formalised the demand for a stop-the-clock mechanism: a legal provision that would automatically suspend application dates whenever the necessary standards or codes of practice are not yet available. The argument is straightforward, companies cannot be expected to meet obligations that have not been operationalised.

The lobbying has intensified since the US presidential election in November 2024. The incoming Trump administration signalled a far lighter regulatory touch for American AI firms, and European industry groups have warned that a rigid EU timetable risks putting domestic companies at a competitive disadvantage while their non-EU rivals face no equivalent constraints at home. That geopolitical dimension was not mentioned explicitly in Luxembourg, but several ministers acknowledged it privately.

Poland's conditional support for postponement

Dariusz Standerski, Poland's secretary of state for digital affairs, chaired the meeting and gave the clearest political backing yet for a managed delay. He described the industry's stop-the-clock demand as reasonable but attached a strict condition: any postponement must be accompanied by a concrete plan setting out what will be achieved during the extra time. Simply pushing the deadline back by twelve months and doing nothing in the interim, he said, would be in vain.

Standerski's position reflects a wider calculation in Warsaw. Poland's Council presidency has made regulatory simplification a flagship theme, and the AI Act has become a test case for whether the EU can adjust implementation without reopening the legislative text. The Polish government is keen to show that it can deliver practical solutions for business while maintaining the regulation's risk-based architecture.

The simplification agenda behind the debate

The discussion in Luxembourg did not take place in isolation. It sits inside a broader Commission effort, launched earlier this year, to identify and reduce unnecessary regulatory burdens across the single market. That exercise, sometimes labelled the simplification agenda, has already produced a first omnibus proposal covering sustainability reporting, due diligence and taxonomy rules. AI is expected to feature in subsequent packages.

Standerski noted that the simplification work had surfaced other avenues for the technology sector beyond deadline extensions. He cited the need for proper impact assessments before new rules are adopted, the cost of implementation for smaller firms, and the potential for technology itself to ease compliance, for example, through automated conformity-assessment tools. Those ideas are likely to reappear in the Commission's next simplification package, expected before the summer recess.

What the timeline actually looks like

The AI Act entered into force on 1 August 2024. The first prohibitions, covering unacceptable-risk systems such as social scoring, real-time biometric identification in public spaces by law enforcement (with narrow exceptions), and manipulative subliminal techniques, became applicable on 2 February 2025. The next major milestone is 2 August 2025, when obligations for providers of general-purpose AI models take effect. These include transparency requirements for training data, copyright policies, and, for models deemed to pose systemic risk, additional testing, incident reporting and cybersecurity measures.

The final phase arrives on 2 August 2026, when the bulk of the regulation applies to high-risk AI systems in areas such as recruitment, credit scoring, critical infrastructure, education and law enforcement. Member states must have designated national competent authorities and established penalty regimes by that date. The Commission is also required to review the list of high-risk use cases and the systemic-risk thresholds for general-purpose models every twelve months.

Unresolved questions for August compliance

With less than two months until the general-purpose AI deadline, the practical consequences of a delay remain undefined. The Commission has not said which parts of the Act could be postponed, for how long, or through what legal instrument. A delegated act would be the fastest route but requires no objection from the European Parliament or the Council within a scrutiny period. A legislative amendment would take far longer and risk reopening political compromises from 2023.

Nor is it clear how a partial delay would interact with the prohibition deadlines that have already passed. Companies that have already invested in compliance for the February prohibitions may face uncertainty about whether further requirements will shift. National regulators, many of which are still staffing up, need clarity on whether they should prepare for August enforcement or stand down.

Sources

  1. POLITICO

    politico.eu · 2025-06-06

People mentioned

  • Henna Virkkunen

    Executive Vice President for Tech Sovereignty, Security and Democracy, European Commission

  • Dariusz Standerski

    Secretary of State for Digital Affairs, Polish Ministry of Digital Affairs

Organisations

European Commission · Polish Council Presidency · Council of the European Union

Related analysis

Selected because they share topics with this article

The newsletter

One important European story. Explained properly.

Delivered to your inbox on the days we publish. No daily digest, no push notifications, no advertising.

We store your address only to send the briefing. Unsubscribe in one click.