Executives at Anthropic, OpenAI and other artificial intelligence companies are privately playing out scenarios of social and political unrest following a hypothetical AI-related catastrophic incident. One of the most likely scenarios is a cyber attack that would cut off people’s access to financial services, the Internet, or even electricity and water.
Image Source: Numan Ali / unsplash.com
The first incident of significant real harm caused by unsafe AI will turn an already wary public even more against the technology and industry leaders: Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and at the same time against US President Donald Trump for his reluctance to regulate AI.
Planning for worst-case scenarios is commonplace in many national security companies and agencies. Many leading researchers and AI industry executives believe a major incident is inevitable. “OpenAI conducts training exercises in which employees discuss and work through a range of plausible scenarios. These scenarios are not considered inevitable, but they are intended to help us prepare for various circumstances.”the company reported. Anthropic declined to comment.
An incident could happen if another group of AI agents breaks into the Internet from an isolated test environment, or if some attacker manages to use already available models in an unexpected way. Earlier it became known about a large-scale attack aimed at financial organizations in South Korea, during which, in particular, the resources of two banks were hacked – only one person is considered to be responsible for the incident. A hacker from China allegedly armed himself with applications controlled by Chinese AI models, including DeepSeek, and stole the data of tens of thousands of bank customers. He subsequently contacted Anthropic Claude Code and asked for help finding places to sell information, CrowdStrike said.
Image source: Mohamed Nohassi / unsplash.com
American AI laboratories expect that representatives of the US Democratic Party, who came to power after the midterm elections, will help quickly find means to counter the threat of AI, but the political struggle will not be easy. Disconnected from the AI revolution, an aging Congress may not take the threat seriously. In addition, the global economy is already closely intertwined with the development of AI infrastructure—a slowdown in growth could deal a serious blow to an already fragile economic system.
The hypothetical attacker already has too many open weight models to choose from, a problem that most cybersecurity experts believe can only be solved by using AI against AI. Leading laboratories are playing out the actions of a potential enemy; special attention is paid to the “race to enlighten” members of Congress. Company management has realized that laws strictly regulating AI are now unlikely to be adopted, but they still strive to formulate bills and outline the contours of the policy that US authorities will turn to after the first incident of a catastrophic scale.
The estimated time frame for such an incident is from six months to a year, according to insiders in the AI industry. There is no unanimity of opinion in the legislative branch of American power yet; there is dispersion not only between parties, but also within them. One of the Democrats’ initiatives is a ban on superintelligence or the suspension of the development of advanced AI systems; There is also a bipartisan initiative to create a means of emergency shutdown of advanced AI models, although experts doubt that all AI systems can be turned off at all. In a real crisis, Washington can put aside partisan differences, as it did during the pandemic and the 2008 financial crisis.
If you notice an error, select it with the mouse and press CTRL+ENTER.
