Anthropic has called on leading artificial intelligence developers to agree a coordinated mechanism that could temporarily slow or pause work on the most advanced AI systems, warning that the field’s current pace could lead to a loss of human control.
In a blog post on Thursday, the company behind the Claude chatbot said that as frontier AI systems become increasingly capable of carrying out complex tasks at high speed, it would be “good for the world to have the option to slow or temporarily pause” further development. Its internal research institute would explore the issue with other organisations and take unspecified steps towards building credible slowdown or pause mechanisms, the company said.
Rival OpenAI took a different line. In a separate statement, the ChatGPT-maker said responsibility for setting limits on AI should rest with elected officials rather than individual companies. “Democratic governments — not private companies acting alone — must ultimately determine the rules, safeguards, and accountability mechanisms,” OpenAI said, adding that no single lab should decide the pace of innovation.
Anthropic warned that AI systems are improving sharply at autonomous software tasks, including coding, and that under current trends and sufficient compute, future systems could potentially design and build improved versions of themselves — a scenario known as “recursive self-improvement”. It said such advances could bring major benefits in science and healthcare but could equally increase the risk of humans losing control over highly autonomous systems.
A coordinated global framework would also be needed to ensure compliance, the company argued, so labs could verify that competitors were genuinely slowing down and that bad actors could not exploit a pause to gain ground.
The proposal lands as wider security concerns intensify. Researchers at the University of Toronto recently demonstrated how AI tools could be used to build a self-propagating “worm” that adapts its hacking strategy as it spreads, with lead researcher Nicolas Papernot warning that even widely available open-source tools could be misused for large-scale attacks. The post also follows a US executive order this week that placed greater responsibility on developers to submit advanced systems for cybersecurity testing before public release.
Calls for a temporary halt are not new. In 2023, a Future of Life Institute open letter signed by Elon Musk and others called for a six-month pause in advanced AI training, but the initiative did not result in industry-wide action. Both Anthropic and OpenAI are reportedly preparing for potential stock market listings, even as they continue to expand rapidly.
The debate matters for Sri Lanka too. The country approved a National AI Policy Framework for higher education in May and produced a sovereign Sinhala large language model accepted to IEEE Access — both of which sit downstream of the frontier capabilities being built by Anthropic, OpenAI and their peers. A similar legal fight between Musk and OpenAI’s Sam Altman — over whether OpenAI is still bound by its original safety-first mission — was decided in OpenAI’s favour by a US federal jury last month.
