Executives at leading technology companies, including OpenAI and Anthropic, are privately modeling scenarios of large-scale civil and political unrest. The potential chaos could be caused by the first catastrophic event caused by artificial intelligence.
Executives at leading technology companies, including OpenAI and Anthropic, are privately modeling scenarios of large-scale civil and political unrest. The potential chaos could be caused by the first catastrophic event caused by artificial intelligence.
This is reported by Axios, citing its own industry sources and company representatives.
Tech leaders say a major incident is almost certain to happen in the next 6 to 12 months. The most likely scenario is a massive cyberattack that could disrupt financial services, the internet, electricity or water supplies. The first instance of serious damage is expected to anger the public and send a wave of anger at developers, including OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei, as well as U.S. authorities for their lack of regulation of the industry.
“OpenAI has conducted preparedness exercises where teams discussed and worked through a range of potential scenarios. These scenarios are not considered inevitable, but are designed to help us prepare for a variety of circumstances,” a company spokesperson said. Anthropic declined to comment.
The risks of using AI for attacks are already growing. In particular, a Chinese hacker recently used tools based on DeepSeek models to hack two South Korean banks and steal the data of tens of thousands of customers, after which he used the Claude Code tool to look for places to sell the stolen data.
The main goal of the current simulations is to prepare a legislative framework for the event of a crisis. Experts understand that it is currently impossible to adopt strict restrictions, but they are preparing proposals that politicians can use immediately after an incident. Possible measures discussed include both radical ones — a ban on "superintelligence" or a freeze on developments, and more realistic ones, such as a mandatory "kill switch" for advanced systems.
