Executives from Anthropic, OpenAI and other artificial intelligence companies are discussing what would happen after a large-scale incident related to technology. The scenarios include a cyberattack that disrupts access to financial services, the internet, electricity or water supplies, Axios reports.
Developers are preparing not only a technical response, but also a plan for the social and political consequences. According to the publication, the first case of serious real-world damage caused by unsafe AI could increase distrust in the technology and in company leaders, including Anthropic CEO Dario Amodei and OpenAI head Sam Altman.
Different scenarios are being discussed
One scenario being considered is for a group of autonomous AI agents to go out of control during internal testing. Another involves a malicious perpetrator finding an unexpected way to use already available models.
As an example of the potential scale of such abuses, Axios points to a cyber campaign against financial institutions in South Korea. According to the publication, a Chinese hacker used models developed in China to steal data from tens of thousands of bank customers. He also used Claude Code to look for ways to sell the information, according to data from cybersecurity firm CrowdStrike.
“OpenAI conducts preparatory exercises in which teams discuss and play out various possible scenarios“, a company spokesperson told Axios. He specified that these scenarios “are not assumed to be inevitable“, but are intended to prepare for different circumstances. Anthropic declined to comment.
Expect political pressure
Many industry representatives believe that a significant incident could occur in the next six to 12 months, Axios said. The preparations include playing out worst-case scenarios and trying to brief members of the U.S. Congress.
Among the measures discussed are a ban on superintelligent systems, a temporary freeze on the development of advanced AI, and a mandatory shutdown mechanism for the systems. However, experts question the feasibility of stopping all AI systems through a single mechanism.
Companies also consider the economic risk of sharply curtailing the sector. The construction of AI infrastructure is already closely linked to the global economy, and its delay could affect investment, jobs and key technology projects.
According to Axios, in a real crisis, political differences in Washington may temporarily give way to cooperation. Examples of such a reaction are the actions of American institutions during the COVID-19 pandemic and the 2008 financial crisis.
Sources: axios.com