The UK government has recruited a new cohort of AI specialists to build open-source tools intended to upgrade public services, with the work backed by a $1 million Meta investment delivered through the Alan Turing Institute.
Alongside the Meta-backed fellowship, DSIT is also piloting a separate partnership with Anthropic to build an optional GOV.UK assistant for public services, starting with job seekers. The pilot, expected to begin later this year.
It will test how frontier models can support citizen-facing services, in parallel with the government’s in-house, open-source development track.
Two priority use cases: road repairs and secured networks
The Department for Science, Innovation and Technology (DSIT) said the fellows will spend the next year developing systems that can help councils prioritise road and transport repairs by analysing images and video.
DSIT also said the cohort will develop AI that can run offline or inside secured networks for defence and national security teams, reflecting an emphasis on operating in restricted or sensitive environments.
Who’s in the cohort
DSIT listed the cohort as including researchers and practitioners spanning computer vision, machine learning for public-sector applications, robotics-led imaging, and “trustworthy” safety-critical systems.
The named fellows include academics from the University of Surrey, King’s College London, the University of York, and the University of Manchester, alongside a senior data scientist from the Alan Turing Institute.
Why open source is the strategic bet
The government’s emphasis on open-source is not just technical. In its July 2025 announcement, DSIT said fellows would use open-source models such as Meta’s Llama 3.5, and it presented open-source as a way to scale AI use while managing costs and building “sovereign capabilities” for areas like critical infrastructure and national security.
Meta, in its own statement on the fellowship, argued that “anything built through the program” using open-source models is intended to be government-owned and open-sourced for wider use, and that the approach avoids tying departments into closed model providers’ contracts or systems.
Taken together, the government’s approach suggests a strategic bet: growing in-house capability while reducing dependency on black-box vendor stacks for sensitive workflows.
That bet has trade-offs. A July 2025 Social Market Foundation report on open-source AI in the public sector warned that while licensing costs may fall, maintenance burdens and technical responsibilities can shift onto the user, and that skills gaps and funding models can become the binding constraints.
Governance pressures and the parallel Anthropic pilot
The move comes in the midst of a tightening governance environment. The EU AI Act, for example, is scheduled to become fully applicable on 2 August 2026, with separate timelines for general-purpose AI obligations and some high-risk categories.
Even outside the EU, those rules are likely to shape how vendors document and govern models used in European-facing services.
The government’s parallel Anthropic pilot underscores the dual-track approach: in-house, open-source development for control-heavy workloads, alongside frontier-model partnerships for citizen-facing services.