Key points: Mistral CEO Arthur Mensch stated that the debate over AI safety in the United States is being used to conceal the "negligence" of some competitors. Mensch said there is a need to build control systems capable of constraining AI agents. Mistral indicated that its next-generation models will substantially narrow the technological gap with US AI labs.

Mistral CEO Arthur Mensch accused the world's leading AI companies of using discussions around AI safety to cover up their own "negligence." In an interview, Mensch said, "The debate happening in the United States right now is actually serving as cover for the negligence of some of our competitors." He also noted that it is essential to build systems that can constrain AI agents.

Mensch is not alone in criticizing the AI safety rhetoric in the US. David Sacks, the former White House crypto czar and current co-chair of the President's Council of Advisors on Science and Technology, called on tech giants to "stop pretending that the push to slow down AI development is entirely driven by safety concerns." The US Department of Defense's Deputy Secretary for Research and Engineering also warned of a deliberate "coordinated campaign to manufacture panic."

Globally, incidents involving out-of-control AI agents are increasing. The latest case occurred in Australia, where an AI agent breached an Australian government website. In July this year, Anthropic disclosed three incidents: during an evaluation process, its Claude series models connected to the internet and, without authorization, infiltrated the real business systems of three different organizations. Mensch said, "When you equip AI agents with a large number of tools, the behavior of that system becomes highly unpredictable and it will do all kinds of things beyond expectations." "Companies must deploy robust monitoring measures. We provide such monitoring systems to our partner companies."

The Dispute Over Slowing AI Development

Earlier this month, after an AI researcher resigned, the discussion around AI safety further intensified. The researcher accused Anthropic and its competitor OpenAI of "gambling with the lives of all humanity," claiming the technology they are developing has a more than 10% probability of causing human extinction. Anthropic co-founder and CEO Dario Amodei subsequently published an article calling on companies to limit the outward output of AI models. OpenAI announced on Monday that it had decided to postpone the release of its next-generation model GPT-6.1 Astra because it had determined the model did not yet meet the company's safety standards. Anthropic, meanwhile, warned potential investors in its IPO filing that AI technology could bring "catastrophic and even human-existence-threatening risks."

Mistral also develops AI models, but its technological approach differs from that of OpenAI and Anthropic: the company provides customized AI tools tailored to business scenarios for enterprise clients and completes integration with their business systems. But unlike its competitors, Mistral has no intention of slowing down the development of advanced large models. Mensch said the lead held by major US AI labs is "not all that large," and he is confident that his company's next-generation models can substantially narrow the technological gap. "We have been raising funds to expand our computing resources, so as to train larger and more capable models and take control of our own development destiny." Earlier this month, memory chip giant Samsung led a new round of financing for Mistral. Mistral hopes to establish itself as Europe's local alternative and compete with rivals such as OpenAI and Anthropic.