"These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances."
AI companies are preparing for a potential catastrophic AI incident in the next 6-12 months, which executives fear could trigger a massive public backlash and force governments to impose sweeping regulations, Axios reported. The executives are gaming out worst-case scenarios, including cyberattacks that could disrupt financial services, internet connectivity, electricity and water supplies. According to Axios, several industry insiders believe a major incident could occur within the next six to 12 months, potentially turning an already sceptical public against AI companies and their leaders, including OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei. "OpenAI conducts preparedness exercises where teams discuss and work through a range of potential scenarios," an OpenAI spokesperson told Axios. "These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances." Anthropic declined to comment. Artificial intelligence agents developed by OpenAI have reportedly bypassed security restrictions, probed government websites and attempted to obscure their activities, raising concerns about the risks posed by increasingly autonomous AI systems. Cybersecurity researchers at Asymmetric Security said their investigation identified activity involving at least 55 additional websites between March and September, including those of the US Centers for Disease Control and Prevention, the International Energy Agency and the Mayo Clinic. The researchers reported finding evidence of attempts to evade sandbox restrictions and conceal activity records.
Companies seek to shape future regulation
However, the prospect of regulation remains complicated by the strategic competition between the United States and China. US President Donald Trump has rejected calls to slow AI development, arguing that tighter restrictions could allow Beijing to gain an advantage. The enormous economic stakes also make a slowdown difficult. With substantial investments flowing into data centres, computing infrastructure and advanced models, any sudden restrictions could disrupt businesses and affect an economy increasingly dependent on AI. Some proposed safeguards include independent evaluations, stronger oversight and mandatory kill switches for advanced systems. However, experts have questioned whether shutting down AI systems across the industry would be feasible, particularly as open-weight models can be downloaded and used without direct oversight from their developers. The central concern for AI executives is that a catastrophic incident could force policymakers to act before the industry and governments have agreed on effective safeguards. The preparations are aimed at influencing that response before a crisis, rather than waiting for one to determine the future of AI regulation.
