Matt Clifford is stepping down as chair of the Advanced Research and Invention Agency after his appointment to a senior international role at Anthropic triggered parliamentary concerns over the governance of a taxpayer-funded research body with significant exposure to artificial intelligence.

Clifford joined Anthropic at the beginning of September as managing director of international affairs, leading the US AI company’s work with governments outside North America. He initially intended to retain the ARIA chairmanship alongside the new job, prompting the House of Commons Science, Innovation and Technology Committee to demand clarity over how conflicts would be identified and managed.

Dame Chi Onwurah, who chairs the committee, said on 7 September that Clifford’s decision to leave had addressed the immediate conflict between ARIA and Anthropic, while adding that “important questions remain” about how the arrangement arose and what assessments the government conducted before it became public.

The episode reaches beyond one outside appointment because Clifford has occupied an unusually influential position in Britain’s recent technology policy. He helped shape the government’s AI Opportunities Action Plan, advised on national AI policy, and has chaired ARIA while the research agency has grown from an experimental funding body into an organisation with more than £400 million committed across programmes and at least £1 billion allocated for its next phase.

ARIA’s independence depends on governance

ARIA was deliberately designed with greater freedom than conventional research funders. Its programme directors can make substantial, long-term bets on technically risky areas without following every process associated with mainstream public R&D programmes, on the argument that scientific breakthroughs are sometimes lost when funding systems optimise too heavily for predictable outcomes.

That independence makes board governance more rather than less important. ARIA is a non-departmental public body sponsored by government, accountable to Parliament and scrutinised by the National Audit Office, while its board is responsible for oversight of strategy and governance. The agency funds universities, startups, individuals, and large organisations and generally does not retain intellectual-property rights or take equity in the work it supports.

Artificial intelligence is already part of that portfolio. ARIA has programmes concerned with AI safety, computing, machine intelligence, and secure interaction between autonomous agents, including a £49.8 million programme seeking tools and research that allow agents to operate safely in untrusted environments.

Anthropic, meanwhile, is one of the frontier AI companies most directly exposed to regulatory, safety, infrastructure, and public-policy questions being shaped by governments. Clifford’s new role puts him in charge of engagement with governments across international markets, making separation between his commercial responsibilities and the leadership of a public research agency difficult to treat as a narrow administrative issue.

The revolving door is becoming structural

Movement between government and technology companies is not inherently improper, and expertise developed in one sector can be useful in the other. Britain’s AI policymaking has relied heavily on people who move between startups, investment, academia, major technology companies, and government, partly because technical and commercial knowledge is concentrated in a relatively small professional network.

Frontier AI has nevertheless reached a point where companies are simultaneously commercial suppliers, infrastructure investors, research organisations, targets of regulation, government partners, and potential recipients of public support. A senior executive dealing with governments can therefore encounter policy questions involving compute, energy, procurement, copyright, safety, research funding, competition, data, and national security rather than a single regulatory domain.

The Commons committee’s concern is consequently about the rules surrounding the overlap as much as Clifford himself. Before his resignation, Onwurah described retaining the ARIA role as a “clear conflict of interest” and asked the government to set out safeguards and boundaries; after the resignation, she said she still expected answers about the conflict assessments undertaken and the mechanisms intended to protect confidence in ARIA’s independence.

Those questions have particular weight because ARIA’s model depends on discretion. Programme directors can allocate large amounts of public money to unconventional research and the agency is meant to take risks that more procedural organisations may avoid, so confidence that decisions are insulated from private interests is part of the institutional bargain that makes its freedom politically sustainable.

ARIA has also moved beyond its startup phase. It says its portfolio now contains more than 100 projects, while current funding ranges from opportunity seeds worth up to £500,000 to programmes with budgets running into tens of millions of pounds. The 2025 Spending Review allocated the agency at least £1 billion to expand its scope and capabilities, increasing both its economic influence and the cost of weak governance.

Clifford’s departure removes the most obvious overlap between the agency and Anthropic, but it does not answer how the dual-role arrangement was initially judged acceptable or what standards should apply when senior figures move between government-backed institutions and companies lobbying those institutions’ political sponsors. Parliament’s request for further detail leaves that issue open, while ARIA’s growing budget ensures its governance will continue to receive scrutiny commensurate with the freedom it was created to exercise.