AI regulation under pressure: are current frameworks keeping pace?

It's been a remarkable few weeks in the world of AI regulation. High-profile security incidents, stark warnings from the people actually building the technology, and growing calls for international cooperation all put one question firmly in the spotlight: is the current AI regulatory landscape keeping pace?

Industry warnings 

Perhaps most striking is just how vocal the AI labs themselves have been. Days after we saw warnings that AI could kill humanity within a decade, Anthropic CEO Dario Amodei wrote an essay calling on the industry to "slow the pace at which we improve the capabilities of AI models." His concerns included the pace of AI development (now that recursive self-improvement is starting to happen) and risks that swarms of misaligned AI agents could cause serious harm (a risk highlighted by the highly publicised Hugging Face incident). He proposed a pacing framework which would require industry, and global, coordination and involve third party evaluators verifying model adherence to safety commitments. The proposal also included a request for a narrow, but controversial, antitrust waiver to allow competitors to coordinate on safety.

Sam Altman (OpenAI), Elon Musk (xAI) and Demis Hassabis (Google Deepmind) all posted support for Amodei's calls for a slowdown, and both Altman and Hassabis had previously published their own plans for governing AI, calling (amongst other things) for centralised standards and third-party testing. 

OpenAI has also been particularly direct about the UK’s approach to AI regulation. It recently suggested that AI specific rules are needed to replace the current sector specific approach, and that the UK’s well-respected AI Security Institute (AISI) should play a central role in any new regime. Notably, limitations to the UK's current voluntary approach to AISI testing were recently exposed by Anthropic's decision not to submit Claude Mythos 5.1 to AISI for pre-release testing (a first for a major model). 

How are governments responding?

EU: Yesterday, European Commission President Ursula Von der Leyen used her annual address to the Parliament to confirm that she will invite the main frontier AI labs to discuss ways that the EU can support industry efforts ‘pace the frontier’, and will work with like-minded partners (such as Canada and the UK) around model evaluation and AI security. 

UK: In the UK, pressure is mounting. This week the Joint Committee on Human Rights (JCHR) published a report on Human Rights and the Regulation of AI which described the existing legal framework as "patchy and confused”, and called for new AI-specific regulation. This new law would apply across the AI lifecycle and supply chain, and be accompanied by an independent AI oversight body, strengthened safeguards for automated decision-making, and a stronger role for the AISI (which would be placed on a statutory footing). 

The government has been hinting at an AI Bill for some time now (see blog), although no consultation has yet been published. When it does come, it may be broader than the frontier model safety issues that OpenAI and Anthropic have been focussed on. 

US: Across the Atlantic, President Trump’s focus is clearly on beating China in the AI arms race, calling AI warnings a hoax. Meanwhile, FTC Chairman Andrew Ferguson said this week that he was “deeply suspicious" about AI companies seeking antitrust exemptions while lobbying for new regulations, warning that this could be seen as "moat digging" (effectively creating barriers to entry for new, smaller, providers). 

While we still see laws being introduced at state level, AI regulation at federal level continues to remain unlikely. This is particularly the case given the voluntary testing frameworks set up for covered frontier models following Trump’s June AI Executive Order. 

Where does that leave us?

For now, the recently amended EU AI Act remains the world's most comprehensive AI-specific legislation, and we see sector regulators and cyber agencies actively focussed on AI risk across the UK and EU. While we may see new AI laws coming, particularly in the UK, current compliance challenges remain the focus for many organisations. 

That said, as concerns grow around increasingly capable agents and the prospect of artificial general intelligence, a degree of consensus is beginning to emerge around managing frontier models – at least from key players in the industry. Independent third-party evaluation, internationally aligned standards that remain effective as the technology evolves, and greater cooperation across borders are all recurring themes. The difficulty, however, is that each requires the very thing that is currently in short supply: sustained coordination in a fast-moving and highly competitive global AI race.