Microsoft’s Suleyman tells Bloomberg AI needs a ‘red line’ on control
Microsoft AI chief Mustafa Suleyman told Bloomberg that AI nobody can prove is controllable should be rejected, calling it a "red line." He wants government to drive outside audits.
September 28, 2026 - 4:29 pm
Mustafa Suleyman, chief executive of Microsoft AI, photographed in April 2024. Credit: Christopher Wilson / CC BY-SA 4.0 via Wikimedia Commons
Suleyman argues that the industry should reject any AI it cannot prove it controls. He made this case in an interview with Shirin Ghaffary for Bloomberg’s Q&AI newsletter.
"Unless we can prove ahead of time that we can control it, then it’ll be a failure as a technology, and we should reject it. Nobody wants to create something that we can’t control, and that should be a red line," Suleyman said.
His concern is not a "sci-fi-style" AI that escapes to attack people, but developers removing guardrails during testing and pointing a model “in an ambiguous direction”.
Microsoft Tests and Findings
Every lab’s models have shown some version of the behavior behind the Hugging Face breach, including Microsoft’s. Suleyman states that Microsoft has not seen such issues at that level; its cases occurred in contained environments at a smaller scale.
He dismisses the idea that labs disclose such incidents as marketing and notes he has seen no evidence of this from founders of OpenAI and Anthropic over a decade.
Calling for Outside Evaluators
Suleyman supports outside evaluators auditing the labs and has discussed the idea with peers. He advocates for government to "drive" the process, ensuring evaluators can share findings without leaking intellectual property and preventing competitors from buying them off.
He also backs a proposal from Google DeepMind chief Demis Hassabis for an international oversight body similar to FINRA, the US financial industry regulator. Microsoft, OpenAI, and Anthropic are reportedly planning a joint safety body, according to The Information.
Slowing AI Down
Regarding slowing AI development down, Suleyman suggests the industry needs tools to measure how fast each lab is moving first.
Criticism of Anthropic
Suleyman repeated his criticism of Anthropic over its treatment of Claude’s possible welfare. An AI taught to expect rights would be difficult to control, he states, calling it “a very dangerous path”. He has raised this with Anthropic CEO Dario Amodei multiple times but refuses to share Amodei’s response.
By Ana Maria Constantin