Anthropic chief wants to slow AI development: Fears software that improves itself out of control

Dario Amodei, who heads the AI company Anthropic, is sounding the alarm about something that sounds like science fiction: AI becoming so good at improving itself that it moves faster than humans can keep up with.

He points to a time frame of six to twelve months in which, according to him, this self-improvement could pose a threat to the internet as a whole.

The solution, he says, is outside control: auditors placed directly at the AI companies, common safety standards across the industry, and global agreements resembling the SALT agreements of the Cold War, when the US and the Soviet Union limited nuclear weapons.

For you as a private AI user or company, this doesn't change anything right now. It's a proposal from one man in the industry, not an adopted law or rule.

Do keep an eye on whether politicians or the EU start taking up the proposal in concrete legislation, though. That would be the time to react; until then you can safely continue your everyday use of AI as normal.