Anthropic co-founder Jack Clark has called for stronger international rules governing artificial intelligence, warning that governments and technology companies have only a limited window to address risks as AI systems become more capable of conducting scientific research and contributing to their own development.
Clark said the ability of AI systems to make scientific discoveries could accelerate innovation while increasing the potential for misuse, including the development of biological weapons and more sophisticated cyberattacks.
Speaking at the 2026 Per Jacobsson Lecture in Bangkok, he argued that the technology’s expanding capabilities made regulation more urgent, not less feasible.
Clark said governments should establish safety-testing requirements, transparency standards and disclosure rules, while working with AI developers to assess risks and determine appropriate limits on the development of increasingly powerful systems.
He also called for restraint in deciding how quickly frontier AI capabilities should advance, including whether limits should be placed on the intelligence permitted in deployed systems.
The challenge is complicated by competition between countries and companies. Governments may hesitate to impose restrictions for fear that competitors will move ahead, while companies face commercial incentives to develop more capable systems.
Clark acknowledged that AI-driven research could produce substantial benefits but said some applications require tighter oversight than others. Biological research, for example, should operate under strict human supervision and transparency requirements, he argued.
AI research itself presents another difficult question because using AI systems to improve future AI could accelerate development while introducing risks that are harder to anticipate.
The discussion also exposed differences in how policymakers assess the threat.
Yan Shen, a professor at Peking University’s National School of Development, called for greater international cooperation on financial-market risks associated with AI agents and the technology’s opaque decision-making processes. She suggested that financial regulation could provide a practical starting point for cooperation in areas where wider political agreement may be difficult.
Nageswaran offered a more cautious assessment, warning that AI’s widespread diffusion could make it harder to establish safeguards comparable with those developed for technologies controlled by a limited number of states.
He said a serious but non-catastrophic accident might force governments to reconsider whether certain AI capabilities should be disabled or placed behind stronger protections.
Clark rejected the suggestion that humanity was doomed but argued that policymakers should act before a crisis forces their hand. He said international cooperation was necessary because the risks and benefits of AI would cross borders, making national measures alone insufficient.
The discussion underscored the central regulatory dilemma: governments want to capture the economic benefits of AI without allowing competitive pressure to weaken safeguards around technologies that could carry significant security and financial risks.
