OpenAI and Anthropic Run Drills for Major AI-Linked Disasters
OpenAI, Anthropic and other tech firms have reportedly conducted private exercises to prepare for severe incidents involving artificial intelligence. The scenarios include cyberattacks on banks, internet access and power or water systems, as well as loss of control over autonomous experimental systems. OpenAI confirmed it holds preparedness exercises, but did not acknowledge a formal joint program with Anthropic or any specific timeline for a disaster.
OpenAI, Anthropic, and other tech leaders reportedly rehearsed privately how they would respond if AI caused or worsened a major crisis. Scenarios included cyberattacks on banks or payments, internet outages, power or water disruptions, misuse of downloadable models, loss of control over an autonomous experimental system, and political pressure for rapid AI restrictions.
Reporting came from anonymous Axios sources. OpenAI confirmed exercises but not a formal joint program with Anthropic or a disaster timeline; Anthropic did not comment. OpenAI rated Astra “critical” for cyber capability, saying it could uncover unknown flaws and devise exploits with suitable tools and access, prompting delayed development phases to test safeguards. Some unnamed industry officials anticipated a serious incident within six to twelve months.
If such drills reflect real risks, banks, payment users, internet-dependent households, and utilities could face disruption, while AI developers may confront tighter scrutiny and slower releases. Government agencies and lawmakers may need clearer incident reporting and coordination, though voluntary commitments could limit consistency. The public may see both greater caution and uncertainty about powerful systems. The impact depends on whether these scenarios remain exercises or become actual events.