NeuralStackRegulators and tech giants are colliding on the cost and tempo of AI oversight, with Meta challenging the UK's enforcement model in court while EU lawmakers move to slow the rollout of the bloc's toughest AI rules. On May 7, Meta filed for a judicial review in London’s High Court, targeting Ofcom’s interpretation of the UK’s Online Safety Act fee and fine structure. The same day in Brussels, EU member states and European Parliament negotiators reached a provisional political deal to delay high-risk obligations under the EU AI Act by more than a year, while adding new restrictions on sexually explicit AI imagery and watermarking requirements.
Meta’s case against Ofcom hinges on a deceptively technical but financially significant question: should compliance fees and potential penalties be calculated on a platform’s qualifying worldwide revenue or only on revenue from UK-regulated services? Ofcom has read the Online Safety Act as authorising fees and fines tied to global qualifying revenue, aligning the regime with the General Data Protection Regulation’s 10% global revenue penalty ceiling. Meta argues that this overreaches, insisting that both fees and penalties should be based solely on the services that are actually regulated in the country where the enforcement occurs.
The stakes for Meta go beyond annual fees, which are expected to be small relative to its overall business. Ofcom has signalled that fees will fall between 0.02% and 0.03% of qualifying worldwide revenue, with a liability threshold starting at £250 million, a structure that would translate into tens of millions of pounds a year for Meta against its multi-hundred-billion-dollar revenue base. The company’s greater concern is penalty exposure: if fines are levied on global group revenue rather than UK-specific income, any major breach could trigger sanctions far larger than anything previously seen from a UK regulator. Meta’s lawyers have also attacked Ofcom’s approach to group-level liability, arguing it is unlawful to base fines on a parent company’s total worldwide income when only particular services are implicated.
Ofcom, represented by lawyer Javan Herberg, has told the court it plans to issue fee invoices in the third quarter of 2026, likely in September, and would need to process refunds if Meta’s challenge succeeds. The legal timetable itself could become a regulatory pressure point: the High Court hearing is listed for October 2026, after invoices go out but before any precedent is set on the core revenue question. Industry players including Epic Games and the Computer & Communications Industry Association may seek to intervene, underscoring how closely other platforms are watching the outcome for signals on their own liability under the Online Safety Act.
In parallel, EU lawmakers spent nine hours on May 7 negotiating a material softening of the AI Act’s implementation calendar for high-risk systems, a category that includes biometrics, critical infrastructure, law enforcement, healthcare and border control applications. Under the provisional agreement, the enforcement deadline for these high-risk systems shifts from August 2, 2026 to December 2, 2027, stretching the runway by roughly 16 months. Machinery already covered by the EU Machinery Regulation will be carved out of overlapping AI Act obligations, a move designed to relieve industrial manufacturers from duplicate compliance duties.
The same deal expands the AI Act’s list of prohibited practices to include AI tools that generate unauthorised sexually explicit imagery, a prohibition explicitly linked to content produced by Elon Musk’s xAI Grok chatbot. That ban is slated to apply from December 2, 2026, marking one of the earliest concrete behavioural restrictions aimed at specific generative AI abuses. Lawmakers also agreed to make watermarking mandatory for AI-generated content starting from December 2, 2026, formalising a technical measure many industry figures have been urging as a basic safeguard against deception, misinformation and non-consensual image creation.
Cyprus’s deputy minister for European affairs Marilena Raouna hailed the changes as a win for EU businesses, arguing that by deferring major obligations and trimming overlap with existing machinery rules, the agreement "significantly supports our companies by reducing recurring administrative costs." At the same time, the compromise has triggered a backlash from civil society: a coalition of 40 consumer, medical and academic organisations has warned that reopening core elements of the AI Act at this stage risks hollowing out the law’s teeth before its most consequential provisions take effect. Their concern is that extended deadlines, even if paired with new bans and watermarking rules, could leave people exposed to high-risk AI deployments without robust safeguards for years.
Why this matters
The UK court fight and EU timetable revision capture the two central tensions in global AI governance: how far regulators can go in extracting financial accountability from tech platforms, and how fast they can realistically demand compliance with complex new obligations. Meta’s push to limit fines and fees to national service revenues, rather than global income, could determine whether the Online Safety Act becomes a powerful deterrent or a more modest regulatory tool. In Brussels, the delayed high-risk AI deadlines give industry more breathing room and cut compliance overlap, but they also raise the risk that the most sensitive AI uses will continue expanding under only partial guardrails, even as sexual imagery and watermarking rules arrive earlier.
Outside the formal legislative process, influential investors are increasingly calling for faster, sharper intervention to control AI-generated content. Billionaire hedge fund manager Paul Tudor Jones told CNBC that the United States is already late on AI regulation and urged mandatory watermarking of AI-generated material coupled with criminal penalties for intentional violations. He framed the issue as a basic need to distinguish "what’s authentically human and what’s not," and said sentiment among AI researchers and model builders has flipped, with a large majority now favouring regulation compared with a minority last year. His analogy to post–World War II nuclear oversight underscored a belief that lawmakers should already have established dedicated AI regulatory structures.
Jones also tied his regulatory alarm to a bullish market view, likening the current AI boom to the acceleration of the internet around 1995 and warning that while the rally may have another year or two to run, the eventual correction could be severe. He disclosed that he has been adding to positions in AI-linked stocks even as he calls for stronger safety rules, a juxtaposition that highlights how deeply intertwined speculative capital and governance debates have become. For policymakers, that combination means any sudden regulatory shock could ripple through not only technology operations but also financial markets that have priced in aggressive AI growth.
The EU’s provisional agreement is not yet law: it still needs formal endorsement from governments and the European Parliament before the delayed deadlines, new bans and watermarking obligations are locked in. In the UK, Meta’s challenge could trigger adjustments to Ofcom’s fee invoices or prompt further legislative clarification on how the Online Safety Act should be applied to multinational platforms. Together with high-profile investor pressure for watermarking and accountability, these moves suggest that the next eighteen months will be defined by a series of legal tests, implementation choices and political compromises that will shape how aggressively AI is policed in its most sensitive applications.