How AI Companies, Alongside the US Government, Have Influenced Decisions on the EU AI Act

AI Act Pull © Neo Institute Europa

Since the European Artificial Intelligence Regulation (AI Act) entered into force in August 2024, its implementation timetable, originally expected to run until 2027, has become the scene of an unusually intense battle for influence in Brussels. On one hand, major US technology companies, led by Meta, Google, OpenAI, Amazon and Microsoft, have deployed substantial lobbying resources to secure a relaxation of the rules. On the other, the administration of President Donald Trump has made European digital deregulation an explicit component of its trade policy, going so far as to threaten the European Union with tariff retaliation. Between the summer of 2025 and the summer of 2026, this dual pressure, both industrial and diplomatic, secured a two-year postponement of the most restrictive AI Act obligations for high-risk systems, enacted through the “Digital Omnibus,” which entered into force on July 27, 2026.

How the EU AI Act was designed to work

The AI Act was designed as a “risk-based” regulation. The greater the potential danger an artificial intelligence system poses to fundamental rights, the stricter the obligations imposed on it. The legislation provided for a phased implementation. Practices considered unacceptable (social scoring, cognitive manipulation) were banned as early as February 2025; obligations for general-purpose AI models (GPAI, such as ChatGPT or Gemini) were due to apply from August 2025 and the most demanding rules, concerning high-risk systems (recruitment, education, healthcare, justice), were scheduled to apply from August 2026.

As early as spring 2025, this trajectory came under growing criticism, expressed on two distinct but complementary fronts: Companies, through traditional lobbying and open letters, and US trade diplomacy, which turned the AI Act into a fully fledged point of transatlantic friction.

The AI Code of Practice and the Stop the Clock campaign

The first concrete confrontation took place around the General-Purpose AI Code of Practice, a voluntary text drafted by thirteen independent experts to help companies comply with the AI Act ahead of the August 2, 2025 deadline. The outcome highlighted a clear divide among major US players.

Meta was the first company to publicly refuse to sign the code. Its head of global affairs, Joel Kaplan, argued on LinkedIn that Europe was heading in the wrong direction and that the text introduced legal uncertainties extending well beyond the scope of the AI Act itself, particularly concerning training-data documentation and copyright-related safeguards. OpenAI and Mistral, by contrast, signed without major reservations, a decision that observers and analysts interpreted as a strategic calculation. Signatories hoped to gain a degree of political recognition from the Commission and easier access to European public procurement markets. Google followed a week later, but accompanied its signature with public reservations. Its chief legal officer, Kent Walker, warned that certain provisions, particularly those relating to copyright or exposing trade secrets, could hamper the development of competitive models in Europe. Microsoft, for its part, indicated that it would probably sign the text.

This episode had a twofold consequence: First, it demonstrated that the balance of power between the Commission and major platforms was not uniform. Some companies opted for tactical cooperation, while others adopted direct confrontation. Second, it served as a stress test for the Commission ahead of the much more sensitive deadline concerning the regime applicable to high-risk systems, scheduled for August 2026.

In early July 2025, around fifty executives from major European companies, including Airbus, ASML, BNP Paribas, Carrefour, Lufthansa, Mercedes-Benz, Philips, Siemens Energy, TotalEnergies and French AI start-up Mistral, sent an open letter to Commission President Ursula von der Leyen entitled “Stop the Clock.” They called for a two-year moratorium on the entry into force of obligations relating to general-purpose models and high-risk systems, citing rules they considered unclear and redundant with other European legislation.

The Commission initially stood firm. Its spokesperson, Thomas Regnier, publicly stated that there would be “no pause, no grace period, no stopping the clock,” emphasising that the legal timetable applied to everyone. But the pressure only increased in the following months. In September 2025, former Italian Prime Minister Mario Draghi, author of a widely discussed report on European competitiveness, himself called for a pause in the AI Act to assess its potential drawbacks, giving an unexpected political endorsement to the private sector’s demands.

This European front, although led by companies from the continent, was operating within a climate largely shaped by positions advocated across the Atlantic. Several of these same companies, including Google, had also called for additional time to comply with the General-Purpose AI Code of Practice a few weeks earlier.

US trade pressure on European digital regulation

At the same time as industrial lobbying, the Trump administration turned European digital regulation into a fully fledged trade-policy issue. As early as February 2025, at the Munich Security Conference, Vice President JD Vance warned Europeans that Washington would not tolerate measures perceived as discriminatory against US technology companies. That same month, Donald Trump signed a memorandum providing for tariffs against countries deemed responsible for “extortion” targeting US companies, explicitly referring to the fines and obligations imposed by the European Union under the Digital Markets Act, the Digital Services Act and, more broadly, the European digital framework in which the AI Act is situated.

This pressure intensified throughout the year. In December 2025, the US Trade Representative (USTR) published a list of nine European companies, including Spotify, Accenture, Amadeus, Mistral, Publicis and DHL, that could face retaliatory measures if the Union did not abandon what Washington described as discriminatory prosecutions and fines against US service providers. In July 2026, twenty-five Republican lawmakers sent a letter to President Trump urging him to launch trade investigations, known as Section 301 investigations, into the Union’s digital rules, particularly targeting the Digital Markets Act but following the same broader logic of challenging the European regulatory framework for digital technologies and AI.

In response to these attacks, some European officials adopted a firm rhetorical stance. European Commissioner for Competition Teresa Ribera, for example, described US pressure tactics as blackmail, insisting that the European regulatory framework was not negotiable. In practice, however, the Commission entered into direct discussions with the US administration about possible adjustments to the AI Act, as part of the broader regulatory simplification initiative launched in autumn 2025.

Technology lobbying in Brussels

Beyond open letters and diplomatic tensions, influence over the AI Act was also exercised discreetly in the corridors of Brussels’ buildings. According to data compiled by Corporate Europe Observatory and LobbyControl, digital lobbying expenditure in Brussels rose from approximately €113 million in 2023 to €151 million in 2025, an increase of more than 33% in two years. The sector reportedly now employs around 890 full-time-equivalent lobbyists, a figure exceeding the number of members of the European Parliament.

Meta alone was reportedly the sector’s largest spender, with around €10 million in lobbying per year, while Amazon is estimated to have spent around €7 million in a single recent year. This investment translates into direct access to decision-makers. During the first half of 2025, major technology companies held 146 meetings with senior European Commission officials, more than one per working day, and 232 meetings with members of the European Parliament.

According to an article-by-article comparative analysis conducted by the same organisations, seven of the eight amendments proposed in the Commission’s digital simplification package essentially reproduced positions that had already been advocated by the technology industry in its lobbying documents. One researcher quoted by the media outlet EU Perspectives went so far as to describe the Commission’s initiative, officially presented as an exercise in regulatory streamlining, as a genuine “radical deregulation agenda.”

The Digital Omnibus and the AI Act timetable

It was in this context that the European Commission presented, on November 19, 2025, its proposal for an “AI Digital Omnibus,” officially justified by practical difficulties. The absence of harmonised technical standards and the inability of several Member States to designate their national competent authorities on time meant, according to Brussels, that applying the rules on high-risk systems according to the original timetable was unworkable.

Negotiations between the Commission, the Council and the Parliament were lengthy and contentious. A first trilogue in April 2026 ended in failure, before a provisional political agreement was reached on May 7, 2026, confirmed by the Council on June 29, and then formally adopted as Regulation 2026/1744, published in the Official Journal on July 24, 2026 and entering into force on July 27, just five days before the original August 2, 2026 deadline.

The final result constitutes a substantial postponement, although less radical than the two-year moratorium initially requested by the companies that signed the “Stop the Clock” letter. Obligations applicable to standalone high-risk systems covered by Annex III of the AI Act have been postponed from August 2, 2026 to December 2, 2027. Those concerning AI integrated into products already regulated under other legislation, such as industrial machinery, covered by Annex I, have been postponed to August 2, 2028. The text also introduces a new ban, welcomed even by digital-rights advocates, targeting AI systems used to generate non-consensual intimate images or child sexual abuse material.

However, several obligations have remained on the original timetable, which qualifies the idea of a total capitulation by the Commission. Bans on unacceptable practices have remained in force since February 2025; obligations concerning general-purpose AI models have applied since August 2025; and the main transparency obligations under Article 50 entered into application on August 2, 2026 as planned, as did the AI Office’s new supervisory powers over GPAI providers.

European reactions to the postponement

This postponement movement was not unanimously supported even within the European institutions. Several Members of the European Parliament, particularly within the Greens and Social Democrats, expressed concern that a circumstantial alliance between the centre and the right could use the Commission’s simplification drive to weaken the broader body of European digital legislation beyond the AI Act itself. Digital-rights organisations such as Amnesty International, EDRi and Access Now denounced an exercise presented as technical simplification but which, in their view, actually represented a substantial rollback of the safeguards originally provided for, particularly concerning the use of personal data to train AI models. The Omnibus package concerning the GDPR, discussed in parallel with the part strictly devoted to AI, thus became a focal point for some of these criticisms, with certain organisations arguing that the “window of political opportunity” created by the competitiveness agenda had been seized by economic actors considerably more powerful than civil society.

Conversely, some smaller actors in the European technology sector, particularly AI-focused start-ups, welcomed the postponement with a degree of relief, arguing that the original timetable, designed at a time when harmonised technical standards did not yet exist, would have exposed them to compliance risks largely beyond their control. This diversity of reactions illustrates the complexity of the issue. This was not simply a duel between Brussels and Silicon Valley, but a multi-level negotiation in which the interests of major US platforms, traditional European industry and local start-ups did not necessarily coincide, even though they temporarily converged around the need for a postponement.

A timeline of pressure and negotiations

To measure the speed at which this combined influence unfolded, it is useful to place the main milestones on a compressed timeline. In February 2025, the Trump administration opened trade hostilities with the memorandum concerning practices deemed discriminatory against US technology companies. In July 2025, the battle shifted to the General-Purpose AI Code of Practice, with Meta refusing to sign and Google and Microsoft signing with reservations, and then to the “Stop the Clock” letter signed by around fifty European executives, initially rejected by the Commission. In September 2025, Mario Draghi’s call for a pause in the AI Act provided additional political support for those advocating a postponement. In December 2025, the USTR raised the stakes by threatening nine European companies with trade retaliation, only days after the Commission had published, on November 19, 2025, its proposal for an AI Digital Omnibus. The first half of 2026 was then marked by difficult institutional negotiations, an unsuccessful trilogue in April, a political agreement on May 7, Council approval on June 29, and finally the entry into force of the final text on July 27, 2026, five days before the deadline originally established for obligations concerning high-risk systems.

This compressed timeline barely eighteen months between the first waves of pressure and the final adoption of the postponement contrasts with the usual slowness of the European legislative process, which had required several years of negotiations before the AI Act itself was adopted in 2024. According to several observers of the issue in Brussels, this acceleration demonstrates the unprecedented scale of the joint pressure exerted by major technology companies and the US administration on the European regulatory timetable.

What this means for EU AI regulation

The sequence of events between 2025 and 2026 reveals a two-pronged mechanism of influence, rarely observed with such clarity in the recent history of European digital regulation. On one side, there was conventional industrial lobbying, measurable in euros and numbers of meetings, which successfully mobilised both the directly affected US giants and part of the European business community concerned about competitiveness. On the other, there was overt US state pressure, combining tariff threats, the naming of European companies and high-level diplomatic initiatives, which transformed a technical AI-regulation issue into a matter of transatlantic trade negotiations.

The European Commission, while maintaining a principled stance and preserving the overall architecture of the AI Act based on risk levels, ultimately gave ground on the most contested issue, namely the timetable for high-risk obligations, while maintaining its course on the law’s most emblematic provisions, including bans on unacceptable practices and transparency obligations. This intermediate outcome illustrates a dynamic that several analysts characterise as a strategic struggle rather than a one-sided balance of power. Brussels retains an important lever, namely access to its internal market, while Washington and industry possess the ability to increase the political and economic cost of strictly enforcing the legislation.

Nevertheless, the precedent thus created could weigh on the AI Act’s next deadlines, particularly as the newly established 2027 and 2028 dates for high-risk systems approach. If the combined mobilisation of technology companies and the US administration succeeded in securing a postponement of several years on this aspect, there is nothing to suggest that the pressure will not be reactivated as the new deadlines approach, in a context in which digital lobbying in Brussels continues to grow and relations between the European Union and the United States remain marked by lasting tensions over digital sovereignty.

The outcome is more complex than a simple European surrender to Washington. The Commission preserved the AI Act’s core framework and its main bans. Through the Digital Omnibus, it also introduced new bans, including on intimate images generated without consent, which industry had not requested. This suggests that well-funded lobbying backed by political pressure can change the pace of regulation more easily than its underlying principles. As the next deadlines approach, the main question is how the AI Act will be implemented. No major EU institution currently expects the law to be abandoned. The debate will instead focus on how quickly the rules take effect, how much room national authorities have to enforce them and how conditions in the European market will shape their application.