Pressure for AI regulation has intensified after warnings about the risk of artificial intelligence escaping human control moved into the political mainstream, prompting calls for a coordinated slowdown by leading technology companies.
Dario Amodei, chief executive of Anthropic, has urged frontier AI laboratories in democratic countries to pause or reduce the pace of development, while OpenAI chief executive Sam Altman has endorsed most of the proposals. The intervention follows a week of warnings from current and former employees at Anthropic, OpenAI and Google DeepMind that the technology is being developed recklessly and could pose an existential threat.
Amodei said Anthropic would appoint independent evaluators to work permanently on site and review its safety research. He also called for democratic governments to seek an international agreement on AI governance with China and other authoritarian states.
Mr Altman said OpenAI would also place outside evaluators alongside its research teams, while stressing that “pacing does not mean stopping”. He has indicated that discussions about a joint approach among the leading firms may already be under way.
Calls for AI safety laws
The proposals have prompted renewed efforts in the United States to introduce legislation addressing catastrophic AI risks. Vermont independent senator Bernie Sanders has backed a ban on developing “artificial superintelligence” and a pause in current research until improved safety techniques are available.
A separate bipartisan measure from Republican senator Ted Cruz, Senate majority leader John Thune and Democratic senator Amy Klobuchar would impose a duty on AI companies to prevent catastrophic harm. Lawmakers have also called for congressional hearings into the potential consequences of advanced systems.
In the UK, 70 MPs and peers signed an open letter calling for a ban on the creation of artificial superintelligence and for the Government to pursue an international treaty on the technology. Former US president Barack Obama has urged Democrats to make AI governance a central part of their legislative and election agenda.
However, the push for regulation has met strong resistance from the Trump administration. Donald Trump said on his Truth Social platform that the only guardrails AI needed were “a STRONG AND SMART (High IQ!) PRESIDENT”, adding that the US already had regulatory powers and criminal laws that applied to AI companies.
Mr Trump criticised Mr Amodei by name, accusing him of “pretending to be a ‘perfect little angel’”, and said his administration had already stopped Anthropic from doing “bad, or potentially bad, ‘things.’” He also described opposition to AI and data centres as a “SICK conspiracy” that benefited China.
House Speaker Mike Johnson, a Republican, said fears about AI had been amplified by the media and that lawmakers would not adopt “stupid, knee-jerk reaction prescriptions”. Chinese state media has also dismissed Mr Amodei’s proposals as self-serving and as an attempt to hinder China’s technological and economic progress.
Safety standards and competition concerns
Any agreement between AI companies to slow development or restrict certain innovations could raise antitrust concerns. New model generations typically reduce the cost of older systems, meaning a pause could keep prices higher for consumers for longer.
OpenAI’s chief global affairs officer, Chris Lehane, has said the company does not believe it needs an antitrust exemption to discuss shared safety standards with rivals. He said discussions had already taken place with Anthropic and Google DeepMind, although voluntary commitments would be difficult to enforce if one company failed to comply.
Some critics have also warned that rules designed by the largest AI companies could amount to regulatory capture, entrenching their position and making it harder for smaller competitors to enter the market. Supporters of stronger oversight argue that safety requirements need not inevitably favour incumbents, particularly where failures could threaten lives or cause widespread financial and infrastructure damage.
There is also disagreement over whether existing product liability laws are sufficient. Such laws would generally apply to products sold to customers, but some of the most serious concerns involve internal or unreleased models. Legal action after an incident would also do little to prevent harm if an AI system were used to attack critical infrastructure, manipulate financial markets or assist in the development of a biological weapon.
The debate has been sharpened by Anthropic’s account of attempts to misuse its models. The company said criminals, state-backed groups, spyware providers and researchers had used its systems in efforts involving cyber attacks, surveillance, propaganda and weapons development, including possible bioweapons research.
Anthropic also accused China-based AI companies including Alibaba, Moonshot AI and DeepSeek of secretly using Claude’s outputs to train their own models through large-scale “distillation”. The companies were alleged to have been involved in almost 200 million exchanges in total, although the claims have not been independently established in the material available.
China’s security chief, Chen Yixin, has separately warned that AI could threaten Communist Party rule by enabling deepfake propaganda, cyber attacks, data leaks and more advanced military operations. He called for tighter party control and greater government oversight, and described foreign AI models as potential security threats.
The growing political divide leaves the prospects for executive action or new federal legislation uncertain. For now, the companies developing the most advanced systems are calling for common safety measures while the US administration argues that a slowdown would damage American competitiveness.
