Trump, Brussels and London set divergent paths for AI regulation in 2025

Three distinct regulatory philosophies are about to collide. On 20 January, Donald Trump returns to the White House with Elon Musk installed as a senior adviser on government efficiency. In Brussels, the European Union's AI Act moves from statute to enforcement, with the first prohibitions taking effect in February. In London, Keir Starmer's government is consulting on a copyright carve-out for AI training that would let rights holders opt out. The result is a fragmented global landscape where the world's largest AI developers face three different rulebooks, each shaped by different political calculations.

Washington prepares an industry-led approach

Artificial intelligence barely featured in Trump's campaign, yet the incoming administration may reshape the sector more than any predecessor. The appointment of Musk, chief executive of Tesla, founder of xAI and an original co-founder of OpenAI, to co-lead a new Department of Government Efficiency alongside Vivek Ramaswamy places a figure with direct commercial stakes in AI at the centre of policy formation. Matt Calkins, chief executive of Appian, told CNBC last month that Musk's presence means "we've finally got one person in the US administration who truly knows about AI and has an opinion about it." Calkins added that Musk has "an unquestioned reluctance to allow AI to cause catastrophic human outcomes" and has warned about such risks long before holding any policy position.

What this translates into remains opaque. No presidential directives or executive orders on AI have been confirmed. The United States still lacks comprehensive federal AI legislation. Instead, a patchwork of state-level bills has proliferated: 45 states plus Washington DC, Puerto Rico and the US Virgin Islands have introduced AI-related legislation. That fragmentation suits large platforms, which can navigate or lobby against fifty different regimes more easily than a single federal standard. Whether a Trump-Musk axis produces a federal framework or simply uses executive power to clear regulatory obstacles for US champions is the central unknown for 2025.

Brussels moves from text to enforcement

The EU AI Act, the world's first comprehensive statutory framework for artificial intelligence, officially entered into force in 2024. But as Shelley McKinley, chief legal officer of GitHub, observed in November: "the next phase of the work has started, which may mean there's more ahead of us than there is behind us at this point." The first provisions become enforceable in February 2025, covering "high-risk" applications including remote biometric identification, loan decisioning and educational scoring. Simultaneously, the EU AI Office, the newly created body overseeing compliance, is expected to publish a third draft of its code of practice for general-purpose AI models.

The second draft, released in December, already exposed friction. It included exemptions for providers of certain open-source models, systems whose weights are publicly available for developers to fine-tune, while imposing rigorous risk assessments on developers of "systemic" general-purpose models. The Computer & Communications Industry Association, whose members include Amazon, Google and Meta, warned the draft "contains measures going far beyond the Act's agreed scope, such as far-reaching copyright measures." The AI Office did not respond to a request for comment. The tension reflects a structural problem: the Act was finalised before the current generation of large language models reached their present capability, and the implementing codes are effectively renegotiating the law's boundaries.

London bets on principles and copyright

The United Kingdom has so far resisted statutory obligations for AI model makers, fearing that premature legislation would choke a sector it wants to attract. Starmer's government has signalled it will introduce legislation, but details remain thin. The expectation in Westminster and among European tech executives is a principles-based regime rather than the EU's risk-based taxonomy. The first concrete move came last month with a consultation on copyrighted content used to train AI models, a flashpoint for generative AI. Most large language models ingest public web data, which includes artwork, journalism and other protected works. Publishers such as the New York Times allege unfair scraping without consent.

The UK proposal would create an exception to copyright law for AI training while allowing rights holders to opt out. Calkins suggested this could make Britain a "global leader" on the issue, noting the country "isn't subject to the same overwhelming lobbying blitz from domestic AI leaders that the US is." That assessment carries weight: the UK lacks a homegrown OpenAI or Anthropic, so its calculus differs from Washington's. But an opt-out regime raises practical questions, how rights holders discover their works in training data, how they exercise the opt-out, and whether the exception survives legal challenge under the Berne Convention. The consultation runs through early 2025; legislation may follow before the summer recess.

Transatlantic friction and the Trump factor

European tech leaders are quietly concerned that punitive EU measures against US platforms could provoke retaliation from a Trump administration. Andy Yen, chief executive of Swiss VPN firm Proton, put it bluntly at Web Summit in Lisbon last November: "[Trump's] view is he probably wants to regulate his tech companies himself. He doesn't want Europe to get involved." The EU has been the most aggressive jurisdiction pursuing antitrust cases against US tech giants, Google, Apple, Meta, Amazon and Microsoft have all faced multi-billion euro fines and behavioural remedies. If Trump treats those actions as trade hostility, the response could be tariffs, investment restrictions or pressure on allied governments to soften their stance.

That dynamic could weaken the AI Act's enforcement before it matures. The Commission may calculate that provoking a trade war over AI regulation is not worth the cost, particularly when the sector's economic weight in Europe remains modest compared to the US. Conversely, a Trump administration that sees AI as a strategic competition with China may want allied regulatory alignment, but on US terms. The EU's insistence on digital sovereignty makes that alignment difficult. The coming year will test whether Brussels can enforce its rules against US companies without triggering a broader rupture.

US-China competition sets the geopolitical frame

Underlying the regulatory divergence is a deeper contest. In his first term, Trump added Huawei to a trade blacklist restricting access to US technology and moved to ban TikTok, owned by China's ByteDance, though he has since softened on the latter. The Biden administration continued and expanded chip export controls, targeting Nvidia's most advanced processors essential for training frontier models. China has responded by accelerating a domestic semiconductor programme, with mixed results. The technological decoupling is real and accelerating.

Max Tegmark, founder of the Future of Life Institute, warns that geopolitical fracturing creates a distinct safety risk. He believes the US and China could each develop artificial general intelligence capable of recursive self-improvement without human supervision. "My optimistic path forward is the US and China unilaterally impose national safety standards to prevent their own companies from doing harm and building uncontrollable AGI, not to appease the rival superpower, but just to protect themselves," Tegmark said in November. The UK hosted a global AI safety summit in 2023 attended by both US and Chinese delegations, but subsequent diplomatic momentum has stalled. If 2025 brings a breakthrough in model capability, the absence of a shared safety framework becomes acute.

Implementation timeline defines the year

The concrete dates in 2025 will shape outcomes more than declarations. February brings the AI Act's first enforceable provisions and the third GPAI code draft. The UK copyright consultation closes in the first quarter, with a government response expected by summer. The US administration's first hundred days, ending in late April, will reveal whether Musk's influence produces an AI executive order or whether the Department of Government Efficiency focuses elsewhere. Meanwhile, the G7 Hiroshima AI process, the OECD AI Principles and the UN's Global Digital Compact all seek convergence, but none has enforcement teeth. The risk is not regulatory chaos, companies are adept at compliance, but that the strictest regime becomes the de facto global standard by default, as happened with GDPR, without the democratic legitimacy of a negotiated agreement.