EU weighs delaying AI Act enforcement under US and industry pressure

The European Commission is preparing to soften the enforcement timetable of its landmark Artificial Intelligence Act, bowing to a concerted lobbying campaign from the incoming Trump administration and a coalition of European industrial giants. Internal documents seen by the Financial Times and reports from specialist outlet MLex reveal that Brussels is weighing a one-year grace period for breaches of the highest-risk AI provisions, a delay in transparency fines until August 2027, and looser post-market monitoring requirements for developers of high-risk systems.

The AI Act, which entered into force in August 2024, remains the world's first comprehensive legislative framework for artificial intelligence. Its most consequential obligations, covering systems deemed to pose serious risks to health, safety or fundamental rights, are not scheduled to apply until August 2026, with a further year's transition for certain categories. The Commission's contemplated amendments would effectively push several enforcement deadlines deeper into the next parliamentary term.

Washington's leverage over Brussels

The pressure from Washington is explicit. Donald Trump, during his campaign and after his election victory, threatened to impose tariffs on countries whose digital regulations or taxes he deemed discriminatory against American technology companies. The EU's AI Act, alongside the Digital Markets Act and Digital Services Act, sits squarely in the crosshairs. For a Commission already anxious about transatlantic trade tensions and the competitiveness of European industry, the threat carries weight.

Thomas Regnier, a Commission spokesperson, acknowledged that "a reflection is still ongoing" on delaying targeted parts of the regulation. He insisted the executive "would always remain fully behind the AI Act and its objectives" and that "it is not for a third country to decide how the EU legislated. This is our sovereign right." The phrasing, defensive, slightly strained, betrays the political difficulty of the moment.

Meta's refusal and the code of practice

The most visible corporate resistance came from Meta. In a blog post earlier this year, Joel Kaplan, the company's chief global affairs officer, declared that "Europe is heading down the wrong path on AI." He argued the Commission's voluntary code of practice for general-purpose AI models introduced "legal uncertainties" for developers and contained measures "far beyond the scope of the AI Act." Meta announced it would not sign the code, a symbolic blow to the Commission's attempt to operationalise the Act's requirements for foundation models before the hard law takes effect.

Kaplan's intervention is part of a broader pattern. US technology firms have spent the past two years arguing that European regulation risks entrenching American dominance by making compliance too costly for European start-ups, while the largest American companies absorb the costs and adapt. The Commission's willingness to consider delays suggests that argument has found traction, at least in parts of the Berlaymont.

European industry joins the chorus

It is not only American firms pushing for breathing room. An open letter signed by the heads of 46 European companies, among them Airbus, Lufthansa and Mercedes-Benz, demanded a two-year pause on the Act's implementation. The signatories framed the request as a test of Europe's "simplification and competitiveness agenda," arguing that innovators and investors needed certainty that the regulatory environment would not strangle deployment of AI in industrial settings.

The letter reflects a genuine anxiety in sectors where AI adoption is still nascent. Aerospace, automotive and logistics companies are experimenting with predictive maintenance, supply-chain optimisation and autonomous systems, but few have moved beyond pilot projects. Compliance costs for high-risk classification, documentation, risk management systems, post-market monitoring, human oversight, are difficult to estimate before the standards bodies finalise technical specifications. A delay, the argument goes, buys time for those standards to mature.

What the Commission is actually considering

According to the Financial Times, the Commission's internal documents outline three principal changes. First, a one-year "grace period" for providers of generative AI systems already on the market before the implementation date, allowing them to adapt practices "without disrupting the market." Second, postponing fines for violations of transparency obligations, such as labelling AI-generated content and disclosing training data summaries, until August 2027. Third, replacing prescriptive post-market monitoring requirements for high-risk systems with non-binding guidance, giving developers greater flexibility in how they track performance once products are deployed.

MLex, which first reported the planned amendments, noted that the proposals could change before their expected release on 19 November. Any formal amendment would then require approval from both the Council of the EU and the European Parliament, a process that could take months and reopen political battles many thought settled when the Act was adopted in March 2024.

Parliament's resistance

Brando Benifei, the Italian Social Democrat who co-led the Parliament's negotiating team on the AI Act, has already drawn a line. "I strongly oppose any 'stop the clock' or delays that would only breed legal uncertainty and leave people exposed to risks the AI act was designed to address proportionately," he said. Benifei's opposition matters: the Parliament must consent to any changes, and the centre-left, Greens and liberals who backed the original text may resist watering down enforcement before a single provision has been tested in practice.

Benifei's critique goes to the heart of the regulatory philosophy underpinning the Act. The legislation was designed as a risk-based framework: the higher the risk, the stricter the obligations. Delaying enforcement for the highest-risk categories, precisely those the Act prioritises, inverts that logic. It also undermines the credibility of the EU's broader digital strategy, which rests on the claim that regulation creates trust and trust enables adoption.

The standards gap

A less discussed but structurally important factor is the state of technical standards. The AI Act relies heavily on harmonised standards developed by European standardisation bodies (CEN, CENELEC) to give companies a presumption of conformity. Those standards are still being drafted. Without them, companies cannot know with precision what compliance looks like. The Commission's own impact assessment acknowledged this gap, but the political timetable, driven by the desire to be first globally, left little room for standards to catch up.

Industry's demand for a pause is, in part, a demand for standards to precede enforcement. The Commission's proposed guidance-based approach to post-market monitoring may be an attempt to bridge that gap without formally reopening the legislative text. Whether that satisfies co-legislators remains to be seen.

Credibility and the global stage

The EU has positioned the AI Act as a global benchmark, the regulatory equivalent of the GDPR for data protection. Countries from Brazil to Canada to South Korea are drafting their own frameworks with one eye on Brussels. If the EU delays enforcement before the law has bitten, it signals that the framework is negotiable, not just in its details, but in its resolve. That perception matters more than any single amendment.

There is also the question of what "grace period" means in practice. If companies know fines will not be imposed for a year, the incentive to invest in compliance now diminishes. The Commission's argument that this avoids "disrupting the market" assumes the market is already moving toward compliance. Evidence suggests many firms are waiting for legal certainty before committing resources.

The coming weeks will test whether the Commission's "reflection" translates into a formal proposal, and whether the Parliament's resistance holds. For now, the AI Act's enforcement clock is still ticking, but the hands may be about to move backwards.