TL;DR
Matt Clifford is leaving the chair of Aria, the government’s frontier research funder, after taking a full-time job at Anthropic running the company’s dealings with governments outside America — the UK included. He had intended to keep both. A select committee chair said the conflict of interest was too plain to manage, and he announced on Monday that he would go on 6 November.
What broke the arrangement
Clifford’s Anthropic appointment surfaced last week, and the original plan was containment rather than departure: he would step back from any Aria business touching Anthropic, under terms the government had signed off.
That did not hold. Chi Onwurah, who chairs the Commons committee covering science and technology, put her objection in a letter to AI minister Kanishka Narayan. Clifford then said he was going so the Anthropic job would not “become a distraction from Aria’s incredible work”, with safeguards applying until he leaves.
Onwurah welcomed it without treating the matter as closed, asking how the situation came about in the first place, what conflict assessments anyone actually ran, and what protects confidence in how Aria is governed. Crossbench peer Beeban Kidron, long a critic of the government’s AI approach, framed it as a trust problem: “we need to see a clear line between tech interests and those of citizens and the nation.”
Not an isolated move
The departure lands on a pattern rather than a one-off. Rishi Sunak has taken roles at Anthropic and Microsoft, OpenAI employs George Osborne, and Nick Clegg spent years at Meta before leaving in 2025. Unlock Democracy’s chief executive Tom Brake has argued the traffic weakens the government’s negotiating hand, because officials carrying sensitive policy knowledge into a lab take the caution with them.
Clifford’s own record is why this one registers. He chaired Aria from 2022, advised Sunak, organised the 2023 safety summit at Bletchley, helped stand up the body that became the AI Security Institute, and served six months as Keir Starmer’s AI adviser from January 2025.
Looking forward
The same day, Geoffrey Hinton backed a private member’s bill outlawing artificial superintelligence, warning that a loss of control over systems cleverer than ourselves could end in human extinction. Both stories point at the same gap. Our analysis yesterday of OpenAI’s mandated safety bars argued the UK’s institutions have influence but no enforcement teeth — and a revolving door is exactly what fills the space where enforcement should be. For UK organisations, the practical read is that AI policy here is still being shaped by people who move between writing the rules and being subject to them.