Could the unimaginable happen within 12 months? Alarm from artificial intelligence giants
A catastrophic incident that would be impossible to ignore: blocked bank accounts, internet collapse, power plants shutting down and water supply interruptions. Such a scenario would bring the dangers of artificial intelligence directly into everyday life, causing anger among citizens. This is precisely the scenario, according to an investigation published by Axios, that the leaders of OpenAI, Anthropic and other artificial intelligence companies are discussing in closed-door meetings.
The most likely scenario is a large-scale cyberattack. It could be caused by autonomous AI systems going out of control during testing, or by criminals managing to use available models in unforeseen ways. The consequences could be devastating, hitting critical infrastructure, public trust in the technology, and the reputation of the companies that develop it.
Simulations for dealing with emergencies are not a new practice. However, what raises more concern, according to Axios, is that many researchers and industry leaders consider a serious incident inevitable. Some of the sources cited by the media believe that an event of this magnitude could occur within the next six to 12 months. An OpenAI spokesperson said that the company conducts preparatory simulations, during which teams examine and deal with various possible scenarios. However, the company emphasized that these scenarios are not considered inevitable, but serve to prepare for different situations. Anthropic, on the other hand, declined to comment.
Such a disaster could cause widespread anger at the leaders of artificial intelligence companies, including Sam Altman and Dario Amodei, but also at US President Donald Trump, due to his reluctant stance on regulating this sector.
According to Axios, companies are particularly preparing for the confrontation with the US Congress. After a serious incident, calls to stop the development of artificial intelligence could become much harder to resist. Among the measures being discussed are the suspension of the development of advanced models and the establishment of a mandatory mechanism for shutting down systems that are out of control. However, implementing such a measure could be extremely difficult. Systems that are already distributed, that can be downloaded and run on different computers, may not be easily stopped with a single switch.
This makes it even more complicated to control technology in case it becomes the source of a crisis with far-reaching consequences.