Executives at OpenAI and Anthropic are preparing for the possibility of a serious AI-related incident that could disrupt banks, internet access, electricity or water supplies.
According to an October 9 report by Axios, their preparations are focusing heavily on how the public and government might react if artificial intelligence causes or helps carry out a large-scale attack.
The companies are reportedly examining worst-case scenarios and considering how to explain the situation to U.S. lawmakers if such an event happens.
These discussions come as more powerful AI systems gain the ability to perform complex tasks and concerns grow over how criminals could use them.
What Kind of AI Disaster Are Companies Preparing For?
One major concern is a cyberattack capable of interrupting services that millions of people depend on every day, ranging from people being unable to use banking services, connect to the internet, or access essential utilities.
Axios reports that the scenarios include AI agents escaping controlled testing environments and criminals exploiting publicly available AI models to carry out attacks. To prepare, executives are reportedly running exercises that examine how the most serious situations could unfold and what governments might do next.
An OpenAI spokesperson confirmed that the company conducts preparedness exercises, but stressed that the scenarios “are not treated as inevitable.”
Why AI-Powered Cyberattacks Are Raising Alarm
Recent cyberattacks have provided a clearer example of what AI-assisted hacking can already involve.
On October 7, cybersecurity company CrowdStrike reported that a suspected financially motivated attacker had used an AI-powered security testing tool called ARTEX during a campaign targeting South Korean financial institutions. Investigators found records linking the activity to AI models including DeepSeek, along with Claude Code sessions in which the attacker sought information about where stolen data might be sold.
The campaign resulted in stolen information, although CrowdStrike said the exact number of affected organisations remained unconfirmed. While it was not the widespread infrastructure failure the companies are preparing for, the incident still demonstrated how AI tools can help one attacker carry out multiple intrusions in a short period.
Why Congress Is Part of the Preparation
According to Axios, much of the planning is focused on preparing to brief members of the U.S. Congress quickly after a major incident. Executives expect that serious harm involving AI could lead lawmakers to demand restrictions, pauses in advanced AI development, or stronger ways to shut down dangerous systems.
However, regulating these systems could be complicated because some AI models can already be downloaded and operated independently outside of the companies that originally developed them. And restrictions could also affect businesses that now increasingly rely on AI tools and infrastructure.
Can Stronger Safeguards Reduce the Risk?
OpenAI has separately called for mandatory national AI safety rules, serious-incident reporting, and international coordination, arguing that risks from advanced models can cross borders.
And so, the reported preparations show that leading AI developers are considering what might happen after serious harm occurs, at least for now. And how better testing, safeguards, and government oversight can reduce the chance of such an incident occurring in the first place.
