OpenAI and Anthropic are developing contingency plans for potential large-scale crises involving artificial intelligence. Industry experts warn that a major incident could disrupt critical infrastructure, including financial services, internet connectivity, and power grids, with some estimates suggesting such risks could materialize within 6 to 12 months.
A primary concern involves AI-powered cyberattacks, where automated tools assist malicious actors in scaling operations. A reported case involved a Chinese attacker using DeepSeek models to steal data from tens of thousands of South Korean bank clients, while also utilizing Claude Code to identify platforms for selling the stolen information.
Beyond technical failures, companies fear that a significant incident could trigger a collapse in public trust and lead to intense political pressure on developers and the US administration. OpenAI confirmed it conducts regular crisis simulation exercises, though it maintains that such disasters are not considered inevitable.
Legislators and industry representatives are currently discussing potential regulatory responses to mitigate these risks. Proposed measures include temporary halts on developing powerful models, restrictions on artificial superintelligence, and the implementation of emergency kill switches for advanced systems, though the effectiveness of such controls remains debated.
