OpenAI and Anthropic Back Binding UK AI Law, Splitting From Rivals
In the UK's frontier AI risks inquiry, the two labs most entangled in London's voluntary safety regime argued for making it law while their rivals argued against it.
A rare split among the leading AI labs
On October 13, executives from OpenAI and Anthropic sat before the UK business, energy and industrial strategy parliamentary committee and delivered a rare joint message: the voluntary safety pledges that frontier AI labs have signed in recent years are no longer sufficient, and the government should move to binding law. The testimony, given as part of the committee's inquiry into frontier model risks, placed the two labs squarely in favor of new legal obligations.
OpenAI senior director for UK and Ireland policy, Amanda Savenko, told the committee that voluntary commitments are no longer enough, and that as AI model capabilities and their safety implications rise, the need for binding regulation grows in step. Michael Sweeney, policy director at Anthropic, made the case from the other direction: rather than asking for a lighter touch, he argued that his company is voluntarily adopting many of the safety case requirements, which means the formal process the government is considering would largely not burden Anthropic as things stand.
The two framings differ but point at the same outcome. Savenko argued from risk: capabilities are advancing faster than voluntary governance can keep pace with, so a new legal framework is needed as an addition to existing safety measures. Sweeney argued from position: a company already invested in safety case preparation would not be put at a competitive disadvantage by codifying those expectations in law, and the industry is ready for the process.
The UK has taken a distinctive path in frontier AI governance, building a voluntary assurance regime around the frontier models deployed on its soil rather than racing to pass a broad AI statute. That approach centers on Safety Case requirements, the Frontier AI Security Centre, and the UK AI Assurance Framework. The committee's inquiry is now testing whether that voluntary foundation should harden into statute, and the hearing agenda made clear the question cannot be decided in isolation from US and EU regulatory developments.
The time to legislate is now, Sweeney told the committee, as MLQ News reported on October 14. That is a notably strong position from a company whose models are scrutinized under the very regime in question. It signals that at least one frontier vendor is confident enough in its own safety investment to ask the state to lock the standards in, converting a framework it can currently choose to honor into one it would be legally obliged to meet. For AI companies operating in the UK, a move to statute would reprice compliance: preparation for a safety case shifts from a strategic decision to a regulatory requirement.
Why the DeepMind and Meta position matters
What makes the hearing newsworthy beyond the substance is the alignment. Google DeepMinds head of policy, Daniel De Freitas, and Meta UK head of policy, Joaquin Catherwood, also testified, and Tech Startups' roundup of the reporting describes them as arguing against a broad new law. That produces a clean split down the industry: OpenAI and Anthropic in favor of binding regulation, Google DeepMind and Meta resisting it.
The alignment matters. The companies most entangled in the UK emerging regime - OpenAI and Anthropic, both deploying frontier models in a way that puts them under the safety case process - want it made into law. Their rivals, less entangled in that specific process, would keep their options open under a lighter-touch framework. That is not a disagreement about AI risk in the abstract. It is a disagreement about which rulebook the industry should adopt and which companies the current voluntary arrangement favors. It is also an early signal of where the AI industry stands as the US and EU debate their own frontier rules, since the same executives and the same competitive dynamics sit at every table.
The committee is still in its inquiry phase. No legislation has passed, and the government has not committed to drafting a new frontier AI law. The hearing is evidence being gathered for a decision taken with international developments in view, and both sides will keep testifying publicly as the US and EU tracks progress. For now, the most durable takeaway is that the industry assumption that frontier governance would remain voluntary is no longer unanimous.
If you build on frontier models, the UK is where binding AI regulation is being actively debated, and two of the largest model vendors are arguing for it, which makes a statutory framework something to watch rather than assume away. If you are an enterprise customer, a hardened regime raises the industry compliance floor, which tends to favor vendors already investing in safety cases. If you follow AI risk, treat the OpenAI-Anthropic alignment against the DeepMind-Meta position as the clearest public evidence yet that frontier AI regulation will become a competitive battleground, not just a policy debate.
Sources: MLQ News; Tech Startups daily roundup.
Corrections: Report an error at corrections@wireshaman.com.