AI leaders back tentative slowdown as critics demand enforceable safeguards

OpenAI, Anthropic and other major AI companies have tentatively backed proposals to “pace the frontier” of artificial intelligence development, according to The Verge. The proposed measures include embedding independent auditors in AI laboratories, regulating domestic labs and pursuing a global slowdown.

The discussions do not amount to a binding agreement. It remains unclear which companies have formally agreed to the proposals, what obligations they would accept and how independent or powerful third-party auditors would be. Auditors could potentially flag or disclose serious safety concerns.

Some researchers and industry figures have welcomed the direction, but said the commitments must be measurable and enforceable. Proposed safeguards include giving auditors access to laboratories’ computing budgets and tying any agreement to actual reductions in development speed or available computing resources.

Critics question whether the initiative would deliver meaningful safety improvements or amount to industry-led regulation. They warn that voluntary rules could disadvantage open-source projects and smaller competitors, while allowing major laboratories to use safety concerns to limit competition or avoid tougher legal requirements.

International coordination is another obstacle. Supporters argue that cooperation is necessary as frontier AI development is increasingly framed as a US–China race. China’s Foreign Ministry has reportedly dismissed calls for a slowdown as “fearmongering”.

The debate has intensified around recursive self-improvement, a hypothetical stage in which AI systems train and improve themselves with little or no human involvement. Researchers have described the prospect as a significant safety concern, but claims about its timing and consequences remain forecasts rather than established outcomes.