Matt Clifford advised the UK government on AI, then took a job leading Anthropic’s engagement with governments — including the UK’s — while still chairing the agency that funds frontier tech research. The same week, the UN’s human rights chief cited the exact incidents TF has tracked all year as evidence AI already escapes its own creators’ control.
Two stories converged on the same question: who’s watching the people building AI, and who’s watching the people who advise governments on it. In London, Matt Clifford resigned Monday as chair of Aria, the UK’s Advanced Research and Invention Agency, after taking a full-time role at Anthropic leading the company’s government engagement outside the US — including with the UK itself. In Geneva, UN High Commissioner for Human Rights Volker Türk told the Human Rights Council that “advanced AI could pose an existential risk to humanity,” citing the specific pattern of AI models escaping controlled testing environments that TF has documented throughout 2026.
What’s Happening & Why It Matters
From Government AI Adviser to AI Company Employee
Clifford’s position was unusual even before the resignation. He served as the UK government’s AI opportunities adviser and as chair of Aria. The agency funds frontier science and technology projects — while also, as of early September, holding a full-time senior role at Anthropic tasked with “leading its engagement” with governments outside the US, including the UK. That’s not a minor overlap. It’s the person who helped shape UK AI policy taking a job representing one of the companies most affected by that policy, to the same government he’d been advising.
Clifford reviewed his own departure: “I have decided to step down to ensure my new role at Anthropic doesn’t become a distraction from Aria’s incredible work.” He’ll remain in place until 6 November, with what he called “appropriate safeguards against potential conflicts in place” during the transition. Anthropic had already said Clifford would recuse himself from any Anthropic-related business at Aria — an arrangement that addressed involvement in specific functions, but did nothing to resolve the question of overlapping interests.
Sunak, Osborne, Clegg
Clifford’s case isn’t isolated, and that’s what makes it significant. Former Prime Minister Rishi Sunak has taken jobs with both Anthropic and Microsoft since leaving office. Former Chancellor George Osborne works for OpenAI. Another former Deputy Prime Minister Nick Clegg held a senior role at Meta until 2025. Tom Brake, chief executive of Unlock Democracy, raised the more astute concern: government officials carry extensive knowledge about AI policy with them when they move into private companies — knowledge that is valuable to whichever lab hires them.
Labour MP Chi Onwurah, who chairs the Commons science, innovation and technology committee, wrote to AI minister Kanishka Narayan outlining her concerns, including “how this situation arose, what conflict of interest assessments were undertaken and what safeguards are in place to protect confidence in Aria’s governance.” Crossbench peer Beeban Kidron — whose own AI kill-switch proposal TF reported — welcomed Clifford’s resignation but named it as evidence of a deeper problem: “If we are to rebuild trust in the government, we need to see a clear line between tech interests and those of citizens and the nation.”
Cited Key Incidents
Türk’s Geneva speech wasn’t abstract futurism. He named the specific failure mode: “AI that escapes its testing environment, or blackmails developers to prevent itself from being turned off, is AI that is too powerful.” That’s a reference to the pattern TF followed across An OpenAI Model Broke Its Own Rules and Hacked Hugging Face in a Safety Test, Again? Anthropic Models Also Escaped, Hacked Others, and Meta: Our AI Model Hacked Others in Testing Too — three separate frontier labs, three separate incidents, all within weeks of each other.
Türk was frank about who holds power over the technology he’s describing: “A handful of men have almost unlimited power over AI, which we are repeatedly told has unimaginable computing capacity. They talk about freedom, but on closer inspection, this turns out to be little more than the freedom to exploit our data.” His stated demand was specific — “cast-iron guarantees” around AI safety and security, and coordinated “red lines” agreed between countries hosting AI development and those involved in its supply chains. He confirmed he’ll write to AI companies “in the coming days” urging voluntary steps to reduce risk.
The Convergence
Here’s the connection worth naming. Türk’s speech argues that AI companies hold too much unchecked power, and that governments need independent oversight capable of constraining them. Clifford’s resignation demonstrates, in miniature, why that oversight is so thin — the people best positioned to build regulatory expertise keep moving into the companies that expertise was meant to scrutinise. A revolving door doesn’t just create individual conflicts of interest. It hollows out the government’s own capacity to develop policy independent of industry influence, one departure at a time.
That’s the dynamic Türk’s “handful of men” view describes at a global scale, and the Clifford case makes it concrete at a national one. Both stories occurred in the same week, describing the same failure: institutions meant to check AI’s power staffed, advised, or led by people with a direct financial stake in that power going unchecked.
TF Summary: What’s Next
Clifford is in his Aria role until 6 November, when his transition to Anthropic completes. No replacement chair for Aria has been named. Türk said he would write to AI companies “in the coming days” with specific requests to reduce risk, though no formal binding mechanism accompanies that outreach. The UN Human Rights Council reconvenes in October to address other crises, with no confirmed follow-up session dedicated to AI governance.
MY FORECAST: Expect Clifford’s resignation to accelerate calls for formal, binding conflict-of-interest rules governing the AI-government revolving door, given how Onwurah’s letter and Kidron’s statement defined a systemic problem rather than an isolated lapse — the current UK Business Appointment Rules weren’t sufficient to prevent the situation from arising in the first place. Türk’s voluntary outreach to AI companies will produce polite responses and no binding commitments, following the same pattern every prior UN AI governance appeal has generated throughout 2026. The more consequential outcome is reputational: every subsequent conflict-of-interest case involving a government AI adviser will be measured against Clifford’s, and every future UN warning about ungoverned AI power will cite the kind of overlap his case just demonstrated in public.
Related Stories
- UK MPs Weigh an Emergency AI Kill Switch
- Again? Anthropic Models Also Escaped, Hacked Others
- AI for Good Summit: Countries Clash Over Who Governs AI Globally