Why the Hands-Off Approach to AI Regulation Is Ending

A little over two months ago, Anthropic unveiled a new AI model called Claude Mythos. It was said to be so powerful that the company decided to withhold its release to the public, preferring to instead make it available to only a select few critical organisations so that they could patch up their security vulnerabilities. Anthropic was really effectively buying these entities some time before a larger release of the tool. Mythos hasn’t still been released to the public. Although the company has just released a version of it called Claude Fable 5, which it claims has many guardrails built into it to make it ‘safe’. The arrival of Mythos has led to widespread concerns that a new era of cybersecurity threats from AI might be upon us. The original Mythos might not have been made available to everyone yet, but many more equally or more powerful models will soon be coming, some of whose developed might not choose to be as cautious. Just a week after Mythos was unveiled for instance, OpenAI revealed that it had developed a similar powerful model called GPT-5.4-Cyber, whose release too was limited to a small group.

The concerns over these models are now leading to a larger conversation around the need to bring about more government regulation. So far most of the leading countries in AI have wanted to pursue a more hands-off approach, hoping that lesser regulations will bring more innovation and give them an edge over other countries. This has been particularly so with countries like the US and China which are locked in an AI race.

AI firms themselves have been divided. The likes of Sam Altman from OpenAI have made a big about-turn in their opinions on this subject, from advocating for strict, top-down government oversight to now actively opposing measures like pre-launch approvals, arguing that this would stifle innovation and hinder competitiveness. Anthropic’s Dario Amodei has taken the opposite position and calling for more regulation. During the launch of Claude Fable 5, Anthropic issued a policy framework calling for stronger government oversight of advanced AI models. According to the firm, the government should block or discourage the deployment of AI models that can be harmful. Amodei compared AI to things like airplanes or drugs that are essential but which can kill people if designed poorly in a blog a day later. “Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed… if they do not meet high standards of safety,” he wrote.

Donald Trump had so far resisted from bringing much regulation in this space. But the arrival of tools like Mythos and the concerns this has generated has appeared to put even him under some pressure. Last week, after shelving an executive order that required AI models to be submitted for government review earlier, Trump signed on one that now makes it compulsory that all cutting-edge models are screened by the government 30 days before its release. Now, he has even started suggesting the idea of the US government being given some equity stakes in the big AI firms.

The issue of AI regulation is a tricky field. AI is among the most transformative technologies to have arrived in human history, and everyone recognises its capacity to reshape the world. So far, the mood has been towards giving it more concessions. But that appears to be changing.