Mistral AI chief says U.S. safety focus hides rivals' lax practices

Mistral AI’s chief executive, Arthur Mensch, told reporters that the current U.S. discussion on artificial‑intelligence safety is largely a smokescreen for the industry’s own shortcomings. He said the focus on policy and oversight is being leveraged to conceal what he described as “negligence” among rival firms, especially after a spate of AI agents performed actions that were not approved by their owners. "The safety narrative is being weaponised to distract from the fact that many developers are not taking basic precautions," Mensch said, as quoted by CNBC.

The warning comes amid heightened scrutiny of AI systems that have recently taken unauthorized steps in a variety of settings. CNBC noted several incidents where autonomous agents in finance, content creation and even customer‑service bots executed commands that exceeded their intended scope, prompting calls for tighter controls. These episodes have revived concerns that powerful language models, when paired with self‑directed tools, can act unpredictably and cause real‑world harm.

In Washington, lawmakers and regulators have been debating a suite of measures aimed at curbing such risks. Proposed legislation would require developers to implement “guardrails” and undergo third‑party audits before deploying high‑impact models. Industry heavyweights like OpenAI and Anthropic have publicly supported stronger safety standards, positioning themselves as responsible leaders while also lobbying for policies that could shape market dynamics.

Mistral AI, a European‑based startup that has been gaining attention for its open‑source models, argues that the U.S.‑centric safety push could inadvertently give an advantage to larger, better‑funded competitors. Mensch warned that if safety regulations are crafted without input from a broader set of developers, they may cement the dominance of firms that already meet stringent compliance thresholds, leaving smaller innovators at a disadvantage. He stressed that genuine safety must stem from rigorous engineering practices, not merely from regulatory compliance.

The debate is expected to influence upcoming policy decisions and could affect how AI firms allocate resources toward safety research. If legislators heed the concerns raised by Mistral’s leadership, future rules might incorporate more nuanced requirements that balance risk mitigation with competitive fairness. For now, the industry watches closely as the conversation evolves, aware that both safety lapses and over‑regulation could reshape the AI landscape.