Microsoft AI's Mustafa Suleyman Says AI Risks Are Real — and Blames Anthropic for Raising Them
In a new interview, Microsoft AI CEO Mustafa Suleyman argued that AI safety threats are genuine while singling out rival Anthropic as a source of escalating risk.
Mustafa Suleyman, the CEO of Microsoft AI, said the dangers posed by advanced artificial intelligence are genuine and argued that rival lab Anthropic is making those dangers worse. His comments came in a wide-ranging interview published as Microsoft released a new position paper on how AI should be built and regulated. The remarks place one of Microsoft's most senior AI executives directly against a competitor at a moment when Washington and Brussels are both weighing fresh rules for frontier models.
Suleyman did not dispute that powerful AI systems carry real risks. Instead, he drew a line between acknowledging those risks and what he characterized as Anthropic's approach to them. Microsoft's own document, published alongside the interview, lays out the company's preferred framework for oversight — a framework that competes with the safety-first posture Anthropic has marketed to regulators and enterprise customers. The timing matters: the two firms are fighting for the same government contracts, cloud partners and corporate buyers.
The public split between Microsoft and Anthropic is the latest flashpoint in a widening argument over who gets to define AI safety. Anthropic, founded by former OpenAI researchers, has built its brand around responsible scaling and has urged lawmakers to treat frontier models as a category requiring special scrutiny. Microsoft, which has poured billions into its OpenAI partnership, has generally favored rules that apply to deployed applications rather than the underlying models. Suleyman's criticism suggests that disagreement is now personal as well as technical.
For enterprise buyers, the practical question is which safety story to trust when both companies are selling the same capability. Microsoft can point to its installed base across Azure and Microsoft 365, where AI features reach hundreds of millions of users. Anthropic has leaned on its safety record to win deals in finance, healthcare and defense, sectors where a model's refusal behavior is part of the pitch. If Suleyman's argument lands, it undercuts the idea that one lab holds a moral edge on safety.
What comes next is whether Suleyman's criticism changes anything in policy. Microsoft's paper is aimed at legislators who are drafting rules now, and the company has a long record of shaping outcomes in Washington. Anthropic has not yet responded publicly to the characterization. The open question is whether this stays a war of words between executives or becomes a formal dispute over which safety standard regulators actually adopt.