The hearing room in Westminster was thick with the kind of quiet tension that usually precedes a hard truth, not a corporate testimony. On one side sat representatives of the world’s most powerful artificial intelligence companies; on the other, a cross-party panel of British MPs determined to find out whether these firms could be trusted to police themselves. What unfolded before the Business, Innovation, Science, and Trade Committee was not just a routine parliamentary evidence session. It was a rare public fracturing of the technology industry’s carefully maintained facade of unity. OpenAI and Anthropic, two of the most influential players in the frontier AI race, told the committee in plain terms that voluntary commitments were no longer sufficient to keep the public safe. They called for binding statutory rules, new laws, and an international convention to govern the development of advanced AI. But their rivals at Google DeepMind and Meta refused to follow. The divide could not have been starker, and it was captured on the parliamentary record for the world to see. Also present were four former employees of these very companies, whistleblowers who have spent months warning that the industry is moving too fast and hiding too much. Alongside them sat senior British civil servants, adding a layer of bureaucratic gravity to an already charged atmosphere. The session was designed to test a simple question: can the people building the most powerful technology in history be trusted to keep it safe on their own? The answer, according to the day’s testimony, was a resounding no from some of the very people who had been doing the building.
Anthropic’s head of Europe, Liam Booth-Smith, was the most direct. When asked whether AI safety now required a statutory framework, he did not hedge. “The rules should be binding,” he said. The company, he explained, had “consistently believed that there should be binding rules on companies and that voluntary commitments are not enough.” This was not a mild preference or a cautious suggestion; it was an admission that the industry’s current approach, built on promises, principles, and unenforceable codes of conduct, was inadequate for the scale of risk involved. OpenAI’s vice president, Tom Duff Gordon, was more measured but ultimately arrived at the same destination. He backed the idea of a statutory framework and went further, calling for an international convention to coordinate oversight across borders. “We don’t think basing everything on voluntary commitments for frontier risk is a good long-term plan,” he said. He also offered a concrete starting point: mandatory incident reporting. If companies were legally required to report serious safety failures, Duff Gordon argued, regulators would at least have a chance to understand what was happening inside the black boxes of frontier AI. Both executives framed their support for regulation not as a burden but as a necessary foundation for the industry’s legitimacy. They seemed to understand that public trust, once broken, is almost impossible to rebuild, and that the only way to earn that trust is to accept oversight that cannot be ignored or quietly abandoned when it becomes inconvenient. Their testimony was striking precisely because it came from inside the industry, from people whose companies would be directly regulated by any new laws. It was a signal that at least some of the people closest to the technology believe the risks are too serious to be left to market forces alone.
But their rivals would not be moved. When Google DeepMind’s Owen Larter was asked directly whether he supported a statutory framework, his answer was evasive in the way of someone trying to avoid saying no while making clear he meant no. “Focus on the standards is what we would be doing,” he said, suggesting that the answer lies not in new laws but in developing technical benchmarks and shared norms. Meta’s Rob Sherman was equally resistant, pointing to the regulators that already exist in Britain and reminding MPs that they were “not writing on a blank page.” In other words, both companies seemed to believe that the current system, however imperfect, was preferable to binding legislation. They argued, implicitly, that regulation could stifle innovation, push research into less transparent jurisdictions, and slow the very progress that could bring enormous benefits. But the committee was not convinced. The hearing exposed a division that has been simmering beneath the surface of the AI industry for years, and now it was on the official record. Liam Byrne, the chair of the select committee, made no secret of what he thought the session had achieved. The committee had set out, he told WIRED after the hearing, “to light up the wide differences in opinions between some of the wealthiest firms on Earth” and to show “that you couldn’t rely on voluntary self-regulation.” That goal was accomplished. The public could now see that even within the small circle of companies racing to build artificial general intelligence, there was no consensus on the most basic question of accountability. If the people leading this technology could not agree on whether to be regulated, how could the rest of society feel safe?
The divide was not merely a matter of corporate strategy; it was a reflection of deeply different philosophies about the nature of AI risk and the role of government. For Anthropic and OpenAI, the dangers of frontier AI are not abstract. They involve the possibility of catastrophic outcomes, from the spread of misinformation to the development of biological or cyber weapons, and they argue that the potential for harm is so great that waiting for a disaster before acting would be unforgivable. Their support for binding rules is a recognition that the industry cannot be trusted to police itself, not because the people involved are malicious, but because the pressures of competition are too intense. When one company pauses to check its work, another rushes ahead; when one agrees to publish its safety tests, another quietly skips them. The presence of the four former staff members turned whistleblowers at the hearing underscored this point. These are individuals who, after leaving their companies, chose to speak out about what they saw on the inside. They have described a culture where safety concerns were sometimes dismissed, where timelines were accelerated for commercial reasons, and where warnings about potential harms were buried in internal memos. Their testimony added a human dimension to the abstract debate, reminding everyone that the decisions being made in boardrooms and research labs will affect real lives. The committee’s decision to include them was deliberate. If the executives were going to make promises, the whistleblowers would be there to provide context about whether those promises had any meaning. The result was a hearing that felt less like a corporate briefing and more like a reckoning.
A few hundred meters away from the parliamentary chamber, another warning was being issued, one that made the urgency of the debate even harder to ignore. Ken McCallum, the director general of MI5, Britain’s domestic security service, used his annual threat update to single out AI as a significant and growing national security concern. The specific danger that most worried him was the potential misuse of biotechnology. “Not yet realized, but the misuse case to which I’d attach greatest weight, is AI’s potential to change the stakes on existing or even new bioweapons,” McCallum said. His words were measured, as befits a spy chief, but they carried an unmistakable weight. The same technology being celebrated for its ability to solve complex problems could, in the wrong hands, lower the barrier to creating pathogens of unimaginable lethality. This was not a hypothetical from a science fiction novel; it was the considered judgment of the person responsible for protecting the United Kingdom from its most serious threats. The proximity of the two events was probably coincidental, but the symbolism was impossible to miss. In one building, executives argued about whether they should be regulated; in another, the head of the security service warned that the stakes of getting this wrong could be measured in lives. The juxtaposition made the industry’s internal squabbles over regulation seem almost petty. If the security services were genuinely worried about AI enabling bioweapons, the question of whether Meta should be subject to binding rules was no longer a matter of business preference. It was a matter of national security.
What the hearing ultimately revealed was that the debate over AI regulation has moved beyond the question of whether it is needed and into the far more difficult territory of how it should be done. The old argument, that AI is too new and too little understood for governments to regulate, is fading. Even those who oppose new laws now frame their opposition in terms of standards and existing institutions, not in terms of leaving the industry entirely alone. The difference is one of degree and enforcement. Voluntary commitments, as Anthropic and OpenAI pointed out, are essentially promises without consequences. They can be made in good faith and broken without penalty. Binding rules, by contrast, create accountability. They allow regulators to investigate, to sanction, and to require transparency. They give the public a mechanism for redress when things go wrong. The fact that two of the four leading AI companies now support this approach is a significant shift, and it may signal the beginning of a broader movement. But it also exposes the fragility of the industry’s collective voice. When the most powerful companies in the world cannot agree on whether to accept oversight, governments are left with a choice: they can act unilaterally, as the European Union has begun to do, or they can wait for consensus that may never come. The British committee’s decision to hold this hearing, and to publicize the divide, was an attempt to push the conversation forward. By putting the conflicting views of OpenAI, Anthropic, Google DeepMind, and Meta on the record, the committee created a lasting artifact of where the industry stood at this critical moment. Future historians may look back at this testimony as the point where the illusion of unified self-regulation collapsed, and where a handful of companies began to acknowledge that the only way to navigate the dangers of AI is together, under the rule of law, with the public’s safety placed above any single corporation’s bottom line. The warnings from MI5 only reinforce that conclusion. The stakes are too high for promises. The time for binding rules is now.
