Anthropic’s Jack Clark urges global AI safeguards within a shrinking window

Anthropic co-founder Jack Clark told policymakers in Bangkok on October 17 that they have only a small number of years to establish safeguards for increasingly powerful AI. Speaking at the IMF–World Bank annual meetings, he paired that warning with a conditional scenario in which US economic output reaches about $44 trillion by 2030, Reuters reported.

Clark called for international cooperation to establish “unusual and serious rules” for the technology. He said AI remained under human control, but warned that faster advances would narrow the opportunity to impose protections. Biological research, he argued, should face the strictest human oversight and transparency.

The assumptions behind $44 trillion

The economic figure was the most striking illustration of the change Clark described. He compared potential US gross domestic product of about $44 trillion in 2030 with just over $30 trillion today—an increase of almost 50% from the current level. He acknowledged that the number “sounds very outlandish, I know.”

That is not a claim that AI alone will add nearly 50% to an otherwise unchanged economy. The comparison is with today’s output, rather than a modeled 2030 economy without AI. Clark said the result assumes continued improvements in the technology; growth on that scale would also require new industries and inventions.

Anthropic’s Economic Scenario Explorer, published in September, supplies the relevant context. Startup Fortune described three paths that depend on how widely AI spreads and how much work it can perform. Anthropic says these are conditional models, not forecasts: they describe outcomes under selected assumptions, not predictions that those outcomes will happen.

$34.1 trillionAI spreads like the internet

The modest scenario puts GDP 1.6% above the modeled 2030 outcome without AI.

$36.3 trillionAI handles half of knowledge work

The substantial scenario assumes AI autonomously performs half of knowledge work, such as tasks involving information and analysis.

$44.4 trillionAI surpasses humans across knowledge work

The extreme scenario assumes AI outperforms people on nearly every knowledge-work task, leaving little alternative work for them.

Scientific assistance and scientific misuse

Clark’s explanation for the potential gains centered on scientific discovery. He described human researchers supported by hundreds or thousands of AI assistants, helping solve mathematical, engineering and biological problems. Drug discovery was one example of work he said could benefit from that assistance.

He expected the acceleration to extend beyond AI development into science as a whole. Addressing hundreds of government and finance officials, Clark said that faster progress would feel confusing and disorienting, bringing both wealth and danger. His warning concerned the pace at which institutions would need to respond, not just the size of eventual economic gains.

The same scientific capabilities could be turned toward harmful ends. Clark specifically raised biological and cyber weapons as possible misuses. That dual use underpinned his demand for especially tight controls on biological research: the capacity to help researchers discover useful things also creates opportunities to misuse science.

Different effects on different jobs

Clark also acknowledged losses alongside the potential expansion. He said AI was already costing jobs in highly exposed fields, including call centers and customer service. Scientific research offered a contrasting case, where he said AI was enhancing employment rather than simply replacing people.

Whether AI creates jobs in entirely new industries remained a major unknown in his account. His employment remarks therefore distinguished between work already being displaced, research work being supplemented and new kinds of employment that might emerge. The last category was a possibility, not an established offset to the losses he described.