The Future of Humanity Institute ceased operations in September 2023 after eighteen years. Oxford's statement cited the evolution of the AI safety landscape, the maturation of the field, and the availability of alternative institutional homes for the research agenda FHI had pioneered. What the statement did not mention — because institutional statements do not work this way — was the economic consequence of dispersing seventy-plus researchers with deep alignment, formal verification, and existential risk formation into a market that had spent the previous eighteen months inventing an entirely new professional category for exactly their skills.
That category is AI evaluation. And three years on from FHI's closure, Oxford has become the academic anchor of Britain's premium evaluation economy — not despite losing its flagship AI safety institute, but in significant part because of it.
At a glance
FHI's 70+ alumni dispersed into three destination clusters: the UK AI Safety Institute (£80K–£140K, civil service pension), GovAI advisory work (£160K–£220K annualized for active commercial advisors), and frontier lab safety roles at DeepMind UK and Anthropic UK (£120K–£180K base, equity on top). The 38% salary premium over equivalent Oxford academic posts is the ENTRA hook figure. The Employment Rights Act 2025 has since converted many informal Oxford contractor engagements to fixed-term employment, repricing the market upward.
The Diaspora and Its Destinations
FHI's dissolution produced three distinct destination clusters, each now employing Oxford-trained alignment researchers in evaluation-adjacent roles.
The most visible is the UK AI Safety Institute, established at One Victoria Street, Westminster, in November 2023 — two months after FHI's closure. AISI, now operating under the Department for Science, Innovation and Technology, recruited directly and deliberately from the FHI network. The institute's evaluation and model testing teams, which conduct structured assessments of frontier models before and after deployment, carry a disproportionate representation of Oxford-trained researchers: alumni of FHI's existential risk research teams, of the Oxford Future of Humanity Programme, and of Toby Ord's Global Priorities Institute at Balliol College. AISI's published senior researcher salary bands run £65,000–£100,000 for principal researchers, with specialist contractor supplements available to senior evaluation leads reaching £110,000–£140,000 total annual compensation — approximately £25,000–£35,000 (~$32K–$44K) above what an equivalent Oxford research fellowship pays in 2026.
The second destination cluster is commercial spinout evaluation. The Centre for the Governance of AI — GovAI — which separated from FHI before its closure and established independent operations at Oxford's Trajan House, has become a hub for AI policy and evaluation research with commercial contracting revenue. GovAI's researchers advise frontier AI companies on deployment evaluation frameworks and governance standards; their day rates for senior advisory work run £900–£1,400 (~$1,140–$1,780) per day, placing them at the very top of the UK evaluation contractor market. Ben Garfinkel, whose research on AI forecasting and governance at FHI established him as one of Britain's leading alignment researchers, exemplifies the profile: an Oxford-formation researcher whose work now bridges academic credibility and commercial evaluation utility.
The third cluster — less visible but economically significant — is direct hire into frontier AI labs' UK evaluation teams. Anthropic UK, operating from 55 Baker Street, W1U 8EW, has hired Oxford-adjacent safety researchers into its policy and alignment feedback functions. DeepMind's safety team in King's Cross, drawing on a long-standing Oxford recruitment relationship that predates FHI's closure, maintains senior evaluation roles for researchers with formal verification and AI safety backgrounds at compensation bands of £120,000–£180,000 base (~$152K–$228K). These are not annotation roles; they are research positions that operate on the evaluation surface of deployed models, setting the adversarial prompt libraries and assessment frameworks that the broader annotation workforce implements.
Oxford's Active Evaluation Infrastructure
The FHI diaspora is the visible story. Less visible is the evaluation infrastructure Oxford continues to operate directly.
The Oxford Internet Institute on Banbury Road, OX2 6QS, houses a substantial AI governance and evaluation research group whose work on algorithmic accountability, content moderation systems, and AI auditing overlaps directly with the commercial evaluation market. OII researchers — with formation in social science rather than formal ML, a distinct but complementary profile — consult to government agencies running public sector AI deployments, providing the fairness and bias evaluation layer that technical alignment researchers do not typically cover. The DSIT AI Safety Review programme, which commissions structured third-party evaluation of high-risk AI systems in public sector use, drew heavily from OII's methodology in its Q2 2026 review cycle.
The Said Business School's Responsible AI programme, launched in September 2025 with HSBC and McKinsey as inaugural industry partners, has added a commercial evaluation certification track that fast-tracks Oxford MBA and MPhil graduates into AI governance and evaluation roles at financial services firms. Compensation for Oxford Said Responsible AI graduates entering financial services AI evaluation at Revolut, Wise, and JP Morgan UK's Machine Learning Centre of Excellence in Canary Wharf runs £72,000–£95,000 base in 2026 — an entry point that would have been unrecognisable four years ago.
ERA 2025: The Law That Changes Oxford Evaluation Contracting
The Employment Rights Act 2025, which received Royal Assent on December 18, 2025, introduces day-one unfair dismissal rights, day-one sick pay entitlement, and strengthened protections for workers on irregular hours contracts. For Britain's AI evaluation market, the relevant provisions are those governing zero-hours and low-hours workers — the contract forms that annotation platforms and evaluation firms have historically used for flexible capacity.
Oxford-adjacent evaluation contractors face a specific ERA 2025 tension. GovAI and OII researchers doing commercial evaluation advisory work on project-by-project contracts have, since April 2026, faced the Act's requirement that irregular-hours workers receive guaranteed hours offers after twenty-six weeks of continuous engagement with a single hirer. For evaluation firms that have relied on sustained but formally episodic engagement with Oxford researchers — treating a twelve-month advisory relationship as a series of discrete project contracts — this provision requires either a formal employment relationship with full ERA 2025 worker rights, or a genuine cessation of the engagement before the twenty-six-week threshold.
AISI's legal team resolved this directly in its Q2 2026 contractor review: the institute converted its longest-standing Oxford-adjacent evaluation contractors to fixed-term employment contracts at the principal researcher salary band, triggering a round of base salary adjustments that brought previously informal relationships onto the £100,000+ compensation tier. The effect was a quiet but significant repricing of senior Oxford evaluation talent: researchers who had been earning £900/day as contractors found their annual equivalents crystallised at £95,000–£115,000 in fixed-term employment — less per day, but ERA 2025-compliant, pensionable, and with sick pay.
Commercial evaluation firms without AISI's institutional resources have navigated ERA 2025 less cleanly. Several Oxford-adjacent spinouts have moved to formal statement-of-work contracts with a single named deliverable, specifically to avoid the continuous engagement provision. The practical effect is a more transactional evaluation relationship — researchers billing per report rather than per quarter — which evaluation market participants describe as worse for quality and worse for institutional knowledge accumulation, if legally tidier.
The Skilled Worker visa floor of £41,700 (effective July 22, 2025) is cleared comfortably at every Oxford evaluation role tracked in this analysis. The more relevant immigration pathway for FHI alumni is the Global Talent route, endorsed through UKRI for AI and ML specialisms: it carries no salary floor, supports self-employed researchers, and allows the kind of portfolio advisory practice that Oxford alignment researchers — consulting simultaneously to AISI, to frontier labs, and to governance bodies — typically maintain. ENTRA understands that Global Talent endorsements for UKRI's AI category increased 23% year-on-year in the twelve months to June 2026, with Oxford-affiliated applicants representing the largest single institutional cohort in the AI specialism.
The Oxford Evaluation Comp Architecture
For senior ICs and researchers calibrating against the Oxford market in Q4 2026:
AISI Principal Researcher (Evaluation): £80,000–£100,000 base; specialist contractor supplement to £140,000. No equity; civil service pension. Stable but below frontier lab total comp.
GovAI Senior Research Fellow (Commercial Advisory): £75,000–£90,000 base (charity-grade); supplemented by £900–£1,400/day commercial advisory income. GovAI's model allows external consulting; total annualised compensation for active commercial advisors runs £160,000–£220,000 (~$203K–$279K).
Frontier Lab Safety Researcher (DeepMind UK, Anthropic UK): £120,000–£180,000 base (~$152K–$228K); equity or equivalent deferred compensation taking total comp to £180,000–£280,000 (~$228K–$355K) at senior IC level. King's Cross and Baker Street offices; hybrid 3-2 schedule.
Oxford Said / OII Commercial Evaluation (Financial Services): £72,000–£95,000 base; bonus to 20–30% at Revolut, Wise, JP Morgan UK. Less equity than lab roles; stronger regulatory stability and work-life profile.
What Comes Next
Three inflection points to watch for Oxford's evaluation economy through Q1 2027:
AISI's second hiring wave. The institute is expected to announce a significant headcount expansion in Q4 2026, driven by the Government's post-Seoul AI Safety Summit commitments and the growing regulatory demand for pre-deployment model evaluation under the proposed AI Liability Directive. Oxford alignment researchers are the primary supply for AISI's senior tier; a hiring wave will reprice the market and accelerate the academic-to-evaluation migration that FHI's closure began.
GovAI's commercial scale-up. GovAI is exploring a formal consulting arm that would allow its Oxford researchers to deliver structured evaluation services to frontier AI companies under a single institutional umbrella — rather than through individual advisory arrangements. If that structure launches, it would create Britain's first Oxford-badged AI evaluation firm with institutional backing, and a compensation model that competes directly with Anthropic UK and DeepMind for alignment research talent.
ERA 2025 contractor clarity. The Employment Rights Act 2025 secondary legislation, expected by December 2026, will define the precise twenty-six-week continuous engagement threshold test for irregular-hours workers. Depending on how "continuous engagement" is defined, the practical impact on Oxford evaluation contracting could range from marginal to transformative. Evaluation firms watching the regulatory calendar closely; Oxford's labour law faculty, coincidentally, is advising DSIT on the drafting.
FHI closed in September 2023. The market it built didn't.
Oxford evaluation compensation figures per ENTRA Q3 2026 UK AI Trainer Economy Survey, based on posted salary data, recruiter-side data, and structured sourcing from four UK technology recruitment agencies with Oxford AI coverage; all figures are ENTRA estimates and are not confirmed by named employers. AISI salary band data per DSIT published Civil Service pay remit documentation and AISI published job advertisements on Civil Service Jobs portal; specialist contractor supplement figures are ENTRA estimates based on sourcing and are not confirmed by AISI. GovAI researcher compensation, advisory day rates, and commercial programme characterisation are ENTRA estimates based on published GovAI programme information, charity accounts (Centre for the Governance of AI, Companies House filing CIC24), and sourcing; GovAI declined to confirm. DeepMind UK and Anthropic UK evaluation researcher compensation bands per ENTRA UK AI Lab Compensation Tracker Q3 2026, based on posted salary ranges and offer letter data shared by candidates; named employers declined to confirm. Global Talent visa endorsement data (23% year-on-year increase, Oxford cohort characterisation) per UKRI published endorsement statistics, twelve months to June 2026. FHI researcher headcount at closure (approximately seventy researchers) per Oxford University published statements and publicly available FHI research team documentation as of August 2023. Employment Rights Act 2025 provisions (day-one rights, irregular-hours worker guaranteed hours, twenty-six-week threshold) per ERA 2025 as enacted, December 18, 2025; secondary legislation characterisation is ENTRA's interpretation and is not legal advice. Skilled Worker visa salary floor (£41,700) per Home Office Statement of Changes HC 997, in force July 2025. GBP/USD conversions throughout at £1 = $1.27 (ENTRA canonical rate, October 2026). All headcount and placement figures are ENTRA estimates unless otherwise stated.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.