Anthropic’s AI-Driven Development Raises New Safety Concerns

Anthropic is using artificial intelligence to help develop more capable artificial intelligence systems, according to a report from the AI safety research organization METR. The report is based on an 11-week investigation conducted inside the company and evaluates Anthropic’s approach to AI research and development.

The findings indicate that Anthropic’s models are improving at roughly twice the pace observed a year ago. The report also found little indication that this acceleration is slowing. These results are being viewed by researchers as some of the clearest evidence so far that AI-assisted development may be increasing the speed of progress in the field.

The report was published on Thursday. Its investigation was separate from another program that gives safety researchers “employee-like” access to Anthropic’s systems. That program was introduced following high-profile hacking incidents during the summer. Anthropic CEO Dario Amodei is shown in connection with the company, but the supplied report does not provide further details about his role in the findings or the methods used.