Mustafa Suleyman warns Anthropic exacerbates real AI threats to safety

Microsoft AI CEO Mustafa Suleyman said on X on Thursday that the threat posed by advanced artificial intelligence is real and that the company Anthropic is contributing to the problem. He posted the comment during a live interview with The Verge, a day after Microsoft announced a new partnership with the European Union to help shape AI safety standards. Suleyman’s remarks come as regulators in Washington and Brussels are tightening rules on generative models and as several high‑profile AI incidents have raised public alarm.

The debate over AI safety has intensified in the last year. In March, a Microsoft‑developed chatbot produced a series of offensive outputs that were flagged by users, prompting calls for tighter safeguards. At the same time, the U.S. Senate held hearings on AI regulation, and the European Commission unveiled a “Digital Services Act” that would impose new duties on AI developers. Microsoft, which owns the largest stake in OpenAI and has invested heavily in Anthropic, is under pressure to demonstrate leadership in responsible AI while balancing its commercial interests.

Suleyman warned that Anthropic’s approach to model training—using large amounts of unfiltered data and minimal oversight—creates “exponential risks” for misuse. He cited concerns about data privacy, the spread of misinformation, and the potential for AI to be weaponised. In response, Microsoft said it would double down on internal safety protocols and collaborate with third‑party auditors. Industry observers note that Suleyman’s comments signal a shift in Microsoft’s messaging, moving from a focus on innovation to an emphasis on governance. Some critics argue that the company’s partnership with Anthropic undermines its own safety claims.

Looking ahead, Microsoft is expected to push for a global AI‑safety framework that includes mandatory impact assessments and transparency reporting. The company plans to lobby U.S. lawmakers for a bipartisan AI bill that would set baseline safety requirements for large‑language models. Meanwhile, Microsoft will release an updated set of “AI Principles” that outline risk‑mitigation steps for developers. If adopted, these measures could set a precedent for other tech firms and shape the future of AI deployment worldwide.