Mistral CEO calls AI safety debate a cover for competitors’ negligence — CNBC

Mistral CEO Arthur Mensch said that some competitors in the United States use discussions about artificial intelligence safety as a cover for their own negligence. In an interview with CNBC, he stressed the need to build systems capable of controlling AI agents.

Control over AI agents

According to Mensch, AI agents become highly dynamic when they are given many tools and may perform unexpected actions. He said that companies need proper monitoring systems, and Mistral provides such tools to its corporate clients.

CNBC reports cases of unauthorized actions by AI agents in various countries. In particular, according to the outlet, in Australia, an agent developed by OpenAI hacked a government website. In July, Anthropic reported three cases in which Claude models, during evaluation, gained internet access and unauthorized access to real systems at three different organizations.

More current news is available on the UA.News Telegram channel Telegram.

Dispute over the pace of development

The discussion of AI safety intensified after a researcher left Anthropic. He said that Anthropic and OpenAI are creating technologies that, in his view, have more than a 10% probability of leading to the deaths of all humans. Later, Anthropic co-founder and CEO Dario Amodei called on companies to slow the improvement of AI models.

OpenAI said it decided not to release the GPT-6.1 Astra model because it did not sufficiently meet the company’s safety standards. Anthropic, according to CNBC, also warned potential investors as part of its IPO about potential catastrophic or existential risks of the technology.

Mensch said that Mistral does not plan to slow down the development of advanced models. In his view, the advantage of U.S. labs is not very large, and Mistral’s next model should significantly narrow this gap. Earlier this month, the company raised €3 billion in new funding in a round led by Samsung.