HMRC awards Quantexa £175m, 10-year deal for AI fraud detection

TL;DR:

• HMRC has signed a £175 million, 10-year partnership with London-based Quantexa to use its Decision Intelligence Platform for sovereign, governed AI at national scale.
• The system will unify HMRC’s fragmented data silos to spot large-scale VAT and payroll fraud, connecting billions of data points to identify high-probability targets that currently go undetected.
• HMRC keeps the data-controller role; automated decisions still require human approval — a deliberate contrast with US-vendor public-sector AI deployments such as Palantir’s NHS work.

HMRC has placed one of the largest UK public-sector AI bets to date, signing a 10-year, £175 million partnership with London-headquartered Quantexa to overhaul how the tax authority handles its data. Chief executive Vishal Marria framed the deal as “a blueprint for how the UK government deploys AI at scale,” with the Decision Intelligence Platform tasked with stitching together fragmented data silos that have historically limited HMRC’s ability to detect complex fraud.

A fraud-detection brief built around connected data

The core use case is fraud. Quantexa says the system will sift through complex webs of shell companies and directors used in large-scale VAT and payroll fraud, connecting billions of data points to identify high-probability targets that currently slip through. By giving HMRC a single connected view of its data, the platform is positioned to support augmented and automated decision-making for both fraud and routine processing — letting the agency redirect resources from administrative checks toward professional criminal networks.

Automated decisions will still need to be approved by human staff. That guardrail matters: spurious AI output has already made its way into UK official documents through lax oversight, and a system operating at HMRC’s scale has limited tolerance for the kind of hallucination that recently forced consultancy firms to retract published work.

Data sovereignty by design

The architecture is the more interesting story for UK SMEs and policy watchers. Quantexa’s technology runs inside HMRC’s own secure environment, with the firm’s staff working with government kept separate from the rest of the business. HMRC remains the data controller for legal purposes. The contrast with deployments like Palantir’s NHS partnership — where concerns have been raised about US firms’ access to UK citizens’ data — is deliberate, and arrives in the same week the new Regulating for Growth Bill named sovereign-AI capacity as a Westminster priority.

The deal stacks on top of HMRC’s recently announced 10-year infrastructure agreement with Amazon Web Services. Quantexa supplies the intelligence layer; AWS supplies the compute — a division of labour that lets HMRC claim both UK-sovereign AI tooling and hyperscaler operational resilience.

Looking forward

For UK AI vendors, the £175 million commitment establishes a procurement template. A long-duration partnership with named data-controller boundaries, transparent and auditable decisioning, and human-in-the-loop sign-off gives Whitehall departments a defensible blueprint they can re-use without going back to first principles each time. Other UK departments shopping similar deals — the DWP’s expanding sentiment-analysis work and the broader Sovereign AI Unit pipeline — should treat the contract structure here as the operative reference point. Whether Quantexa’s fraud-detection numbers come in at the scale the deal value implies will be the test that determines whether the blueprint sticks.