OpenAI and Anthropic prepare for a major AI-driven crisis

OpenAI and Anthropic executives are conducting secret crisis simulations to anticipate political and public reactions to a likely AI incident within the next 6-12 months. The biggest concern is scenarios in which AI agents attack critical infrastructure or malicious actors use existing models for cyberattacks.