Behind closed doors, executives at leading artificial intelligence companies are planning what to do after the first major disaster caused by the technology. They consider a cyberattack the most likely scenario—one that could cripple banks, the internet, or even electricity and water supplies. The industry expects such an event to trigger a wave of public outrage and political pressure. Companies are therefore already trying to influence future rules.
An Incident Within a Year
Numerous people in the artificial intelligence (AI) industry told Axios that they expect a serious incident within six months to a year. Modeling worst-case scenarios has long been standard practice at many companies, and the Pentagon has conducted war games for decades. What is different is that many leading researchers and industry executives consider such a disaster inevitable.
According to the publication, the first instance of real harm from dangerous AI would turn an already wary public further against the technology. Industry leaders, including Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, would face a backlash. So would U.S. President Donald Trump, who has been reluctant to regulate AI.
An OpenAI spokesperson explained that its teams work through various scenarios as part of preparedness exercises but do not consider any of them inevitable. Anthropic declined to comment.
An Attack on Korean Banks
The search for someone to blame will begin regardless of the scenario. An incident could involve a swarm of autonomous AI agents escaping a test environment or a malicious actor finding an unexpected use for publicly available models. We will examine these scenarios and the broader situation in the AI industry in more detail in a video already in production for the Universe Space Tech YouTube channel.
A recent campaign against South Korean financial institutions, described by cybersecurity company CrowdStrike, shows how much damage a single person can cause. A hacker believed to be from China used Chinese AI models, including DeepSeek, to steal data belonging to tens of thousands of bank customers. Two banks were reportedly breached, and the hacker turned to Anthropic’s Claude Code for help finding places to sell the stolen information.
Restricting such uses is difficult. Many AI models are released with open weights, meaning anyone can download them and use them maliciously. Most cybersecurity specialists believe that the only way to counter this is with other AI systems.
Betting on Congress
The companies are working on the assumption that Democrats will control both chambers of Congress after the midterm elections and will try to quickly curtail AI development. However, the companies believe Democrats would face serious obstacles. Congress, dominated by older politicians, has a poor understanding of the new technology, making it difficult to pass laws that actually work. The global economy is also already closely tied to the construction of AI infrastructure, so slowing it down could cause significant economic damage.
Among the most stringent Democratic proposals are a ban on superintelligence and a pause in the development of advanced AI systems. Other proposals have support from both parties and even parts of the industry, such as a mandatory emergency shutdown switch for powerful models. Experts, however, doubt whether shutting down all AI systems is possible at all.
The companies understand that no AI regulation bill will pass right now. They are therefore focusing on educating members of Congress to influence the rules Washington may turn to after the first disaster. In a genuine crisis, politicians can set aside partisan differences. One Democratic aide pointed to cooperation during the COVID pandemic and the 2008 financial crisis.
