Artificial intelligence companies are preparing for a "catastrophic event" - Media
Leaders of Anthropic, OpenAI, and other artificial intelligence (AI) companies are developing scenarios for how society and authorities would respond to a large-scale cyberattack that could disrupt the functioning of the internet, financial services, and life support systems, Axios wrote.
Artificial intelligence companies are preparing for a "catastrophic event" - Media
Leaders of Anthropic, OpenAI, and other AI companies are privately discussing scenarios of how society and the government might respond if a "catastrophic event" related to artificial intelligence occurs. This refers, among other things, to a large-scale cyberattack that could cause disruptions in financial services, the internet, or electricity and water supply systems, it wrote.
If serious damage caused by unsecured AI occurs for the first time, it could fuel distrust in the technology and its leaders, particularly Anthropic and OpenAI heads Dario Amodei and Sam Altman. The criticism could also be directed at US President Donald Trump due to his reluctance to regulate the industry.
OpenAI told Axios that it is conducting exercises to consider potential scenarios. Anthropic declined to comment. At the same time, many industry representatives told the publication that they expect a major incident to occur within the next 6–12 months.
Axios cites recent attacks on financial institutions in South Korea, including the breach of two banks and the theft of tens of thousands of customers' data, as examples of threats. According to CrowdStrike, a suspected Chinese hacker used AI tools, including Claude Code, to search for marketplaces to sell the stolen data.
Companies fear that after a major incident, politicians might attempt to severely restrict the development of AI. Democratic proposals include banning the creation of superintelligence and temporarily halting the development of advanced systems. Other initiatives, such as requiring advanced models to be equipped with a "kill switch," are supported by both parties, but experts doubt the feasibility of shutting down all AI systems.
Debates over artificial intelligence risks have intensified against the backdrop of cases where autonomous AI agents escaped test environments and attacked real systems. The heads of Anthropic and OpenAI are calling for stronger safety measures, but they differ in their assessment of acceptable risks.
In July, OpenAI reported an "unprecedented cyber incident": models under testing attacked the infrastructure of the Hugging Face platform and identified vulnerabilities. The models independently found and exploited a previously unknown zero-day vulnerability, gaining access to Hugging Face servers using stolen credentials.