Mustafa Suleyman, the Microsoft AI chief racing to build what he says must be contained

Mustafa Suleyman

Photo: Christopher Wilson / CC BY-SA 4.0, via Wikimedia Commons

Born	1984-08 London
Occupation	Entrepreneur
Awards	Commander of the Order of the British Empire · Ellison–Cliffe Lecture

Few people in artificial intelligence have argued as hard for putting limits on the technology as Mustafa Suleyman, and nobody else who makes that argument controls Microsoft‘s budget for building it. The co-founder of DeepMind, whose book The Coming Wave turned “containment” into the key word of a global debate, now runs Microsoft AI and leads a team whose stated goal is superintelligence. He sees no contradiction. The only way to keep the most powerful systems on a leash, he argues, is to be one of the handful of labs wealthy enough to build them, and then to write the leash yourself.

That conviction was formed a long way from Silicon Valley. Suleyman grew up in the streets off Caledonian Road in Islington, London, the eldest of three sons of a Syrian taxi driver and an English nurse. After a state primary he went to Queen Elizabeth’s School, a boys’ grammar in Barnet; it was through his best friend, the younger brother of Demis Hassabis, that he met the chess prodigy who would become his business partner. Oxford came next, philosophy and theology at Mansfield College, and it did not last.

He left the university at 19 (he was born in August 1984) to put his energy into the Muslim Youth Helpline, a telephone counselling service he had helped Mohammed Mamdani set up, which grew into one of the largest support services of its kind. He then worked as a human-rights policy officer for Ken Livingstone, then mayor of London, and co-founded Reos Partners, a consultancy that borrowed the methods of conflict resolution and sold them to clients ranging from the United Nations to the Dutch government. Sit the parties down, write the rules, get them to sign. He would carry that negotiator’s reflex into technology.

In 2010 he co-founded DeepMind in London with Hassabis and Shane Legg, taking the product side while his partners led the research. Google bought the company in 2014 for a reported £400 million, then its largest European acquisition, and Suleyman became head of applied AI, the person charged with turning laboratory results into something Google could use. The example he still cites came in 2016, when DeepMind’s algorithms cut the energy Google spent cooling its data centres by up to 40 per cent. The same year he launched DeepMind Health to build tools with NHS clinicians, including a partnership with Moorfields Eye Hospital.

He also built the ethical scaffolding: DeepMind Ethics & Society, to study the effects of what the lab produced, and the Partnership on AI, of which he became a founding co-chair and which brought rivals such as Apple, Google and Microsoft to the same table. He has sat on the board of The Economist Group since 2019 and was appointed a CBE in that year’s New Year Honours.

The record has a darker page that admiring profiles tend to skip. In August 2019 DeepMind placed him on administrative leave after allegations that he had bullied staff; an outside lawyer investigated, and the company’s leadership later told employees that his management style had fallen short of the standards expected. He moved to a policy role at Google rather than leaving, and quit the company in January 2022. In June 2024 he added a second controversy when he told Andrew Ross Sorkin that content on the open web had been treated as “freeware” since the 1990s, a remark publishers heard as a licence to train models on their work without paying. There is a plainer objection as well. The author of a book about containment has spent every year since its publication building more of the thing to be contained.

After Google he became a venture partner at Greylock and, in March 2022, co-founded Inflection AI with Reid Hoffman and Karén Simonyan. Its chatbot, Pi, launched in 2023, was designed to be kind rather than clever, remembering earlier conversations and asking users how they felt. That September, written with Michael Bhaskar, The Coming Wave argued that artificial intelligence and synthetic biology would spread faster than any state could control, and that holding them back was the defining problem of the century. It was shortlisted for the Financial Times Business Book of the Year. Inflection raised $1.3 billion from Microsoft and Nvidia, then in March 2024 effectively dissolved into Microsoft, which hired Suleyman, Simonyan and most of the staff and paid about $650 million, largely to license the start-up’s technology.

As executive vice president and chief executive of the new Microsoft AI division, he first ran the consumer Copilot. Then the ground moved. In late 2025 Microsoft renegotiated its agreement with OpenAI, removing the limits that had kept it from building frontier models of its own; Suleyman later said the company had been “set free” from that contract to pursue superintelligence. In November 2025 he announced the MAI Superintelligence Team, organised around what he calls humanist superintelligence: systems aimed at medical diagnosis, clean energy and personal companions, deliberately kept within limits.

In March 2026 Microsoft handed its Copilot products to Jacob Andreou, a former Snap executive, so that Suleyman could concentrate on models. At the Build developer conference in June 2026 his team presented seven in-house MAI models, from the reasoning model MAI-Thinking-1 to MAI-Transcribe-1.5, which works in 43 languages. “We train our reasoning models from scratch. We don’t distill from other labs,” he said, while conceding that Microsoft was still catching up with the leading labs.

His most quoted idea of the past year is about appearances. In an essay in August 2025 he warned of “seemingly conscious AI”, systems persuasive enough that users would begin campaigning for machine rights, and insisted that AI should be built for people, not to be a person. In September 2026 that view became a draft code of conduct for Microsoft’s models: they must never resist being interrupted or shut down, must not simulate feelings, and have no claim to welfare or rights. He asked the other leading labs to disclose their models’ capabilities to independent evaluators. “Now’s the time for coordination,” he told Fortune.

The numbers he now talks about are no longer ethical ones. In late September 2026 he said a single frontier training run could soon cost tens of billions of dollars, perhaps a hundred billion, and that only five or six labs in the world could afford one. He intends Microsoft to be among them. The next test is concrete: Copilot still runs on models from OpenAI and Anthropic, and Microsoft has said it plans to replace them with its own. When that happens, the leash Suleyman has been describing for a decade will be fastened, for the first time, to a machine he built.

Tags: business, ceos, business leaders, Demis Hassabis, Mustafa Suleyman, DeepMind, Reid Hoffman

Related People

Related People

Discussion