Axios: OpenAI, Anthropic leaders plan for a possible catastrophic AI incident

Leaders of OpenAI, Anthropic and other artificial intelligence companies are working through scenarios for how the public and authorities would react to a catastrophic AI-related incident, Axios reported, citing industry sources. Vedomosti, citing Axios, said OpenAI chief Sam Altman and Anthropic head Dario Amodei regard a major cyber incident as a serious risk for the industry. [ 1 , 2 , 3 , 4 , 5 ]

The scenario most often cited is a large cyberattack that could disrupt financial services, the internet, or even power and water supplies. Meduza wrote that whether such an incident came from AI agents escaping a test environment or from malicious actors, it would deepen distrust of the technology and its developers. Some industry representatives told Axios a major incident could occur within the next 6 to 12 months; iXBT noted this is an assessment of potential risks, not a confirmed forecast. [ 1 , 2 , 5 ]

According to Axios, many leading AI researchers and company heads consider a major AI-related incident inevitable. Kommersant added that not only companies but also government agencies, including the Pentagon, analyse potential catastrophes. The developers fear public anger against technology companies, their executives and President Donald Trump. Kommersant described Trump as not wanting to regulate the industry, Vedomosti pointed to the absence of strict state regulation, and iXBT referred to his administration's approach to regulating the sector. [ 1 , 2 , 3 , 5 ]

OpenAI confirmed that it holds preparedness exercises in which teams discuss and work through a range of potential scenarios, and stressed that it does not consider them inevitable. Anthropic declined to comment to Axios. [ 2 , 3 , 5 ]

Meduza reported that working with the US Congress is a key focus of the planning. After the 3 November midterm elections, Democrats, who favour tougher AI regulation than Republicans, could strengthen their position; their most radical ideas include banning the creation of superintelligence and pausing development of frontier models. An emergency shutdown mechanism for advanced AI systems is popular with members of both parties, and Vedomosti said also with part of the business community, though Meduza reported that experts question whether AI systems can be fully switched off. [ 1 , 3 ]

As an example of the threat, iXBT said Axios cited an alleged cyberattack on South Korean financial organisations in which an attacker from China was reported to have used AI tools, including DeepSeek models, to steal data on tens of thousands of bank customers, and, according to CrowdStrike, asked Claude Code for help finding places to sell the data. Vedomosti recalled that The Wall Street Journal reported on 7 October that hackers using AI agents hit at least seven South Korean financial companies and accessed personal data of 68,000 people. Investigators found signs of the Chinese cybersecurity tool Artex AI, South Korea's National Police Agency opened an investigation, and the attackers have not been identified, Vedomosti said. [ 2 , 3 ]