OpenAI and Anthropic Prepare Behind Closed Doors for a Major AI Incident

According to an Axios report dated October 9, executives at major AI companies, including OpenAI and Anthropic, are working behind closed doors through scenarios for a serious incident involving artificial intelligence. The discussions concern the technical, political and public consequences of a possible failure.

What They Are Drilling

Participants are considering cyberattacks capable of disrupting banking systems, the internet and communications, as well as energy and water supply. The exercises test possible failures through red teaming, assess threats from autonomous AI agents, and prepare responses to state interference.

OpenAI confirmed that it runs exercises to prepare for various risk scenarios. The company stressed that this does not mean a catastrophic incident is considered inevitable. Anthropic declined to comment on the reports.

According to Axios, some industry figures believe a major AI incident could occur within the next six to twelve months. These assessments remain forecasts, not established calculations.

A Recent Case

The report gives a recent example. The campaign targeted financial institutions in South Korea, including reports of breaches at two banks. According to CrowdStrike, the attacker is said to have used Chinese AI models, including DeepSeek, in operations with stolen customer data, as well as Claude Code to look for possible markets for the stolen information.

Political Consequences

Executives are also preparing for the political consequences of a major failure. Such an incident could strengthen demands for stricter oversight, mandatory safety mechanisms, or restrictions on the development of advanced systems. Discussions include, in particular, emergency shutdown of powerful AI systems and temporary pauses in certain areas of development.

Implementing such measures is not easy. Open-weight models can run outside centralized infrastructure, which makes a complete shutdown difficult. Economics also complicates intervention: investment in AI sustains a significant part of technology infrastructure development, creating competing priorities between growth and safety.

Translation editor: Салтанов А.