Verifoxx Wins Sovereign AI Phase 2 Funding to Advance AI Assurance Capability

20 July 2026, London, UK 

Verifoxx has secured £525,000 in follow-on funding from the Sovereign AI Pipeline competition to advance SentinelOps, Verifoxx’s sovereign AI runtime assurance capability for AI systems operating in sensitive, zero trust environments.

The project forms part of a £9 million investment by Innovate UK  in demonstrator stage frontier AI technologies that support the UK’s AI sovereignty objectives.

SentinelOps Phase 1 was one of just 25 projects funded from more than 1,100 applications. In Phase 1, Verifoxx proved that AI workloads could be checked locally, at the point of execution, without exporting raw operational data.

Building on that success, Phase 2 will progress SentinelOps from a TRL 4 proof of concept to a TRL 6 deployable assurance capability in 12 months. It will create a lightweight assurance layer that can be added to AI systems to show whether they are behaving within agreed behavioural and policy boundaries. It represents a new-to-the-world capability in runtime behavioural assurance across interacting autonomous systems.

SentinelOps addresses a major limitation: conventional AI assurance relies on cloud access and shared operational data, which are often unavailable. This makes it difficult for organizations to independently verify that an AI system operated within its expected boundary. This challenge intensifies across distributed and multi-agent AI systems, where multiple autonomous systems collaborate and make decisions without any single operator having a complete view.

SentinelOps gives organisations an independent way to assess whether an AI system remained within its expected behavioural scope while it was running. It checks behaviour at the point of execution and produces a trusted assurance result, even when systems are disconnected and sensitive operational data cannot be shared.

Testing across distributed AI systems, edge devices, and secure computing environments will evaluate SentinelOps' capability to detect compromised behaviour, differentiate legitimate updates from harmful changes, identify system disagreements, and maintain functionality during network disruptions. 

Bala Subramanyan, CTO and Co-Founder of Verifoxx, said:

“Phase 1 showed that we could assess AI behaviour locally, inside constrained environments, without exporting raw operational data.

“Phase 2 turns that proof of concept into a working demonstrator capable of providing runtime assurance for individual AI workloads and supporting trust across distributed and multi-agent AI systems.

“Our goal is to deliver a practical assurance layer that can identify behavioural anomalies, compromise and coordination failure while AI systems are operating, even when networks are disrupted and information is incomplete.”

This funding extends Verifoxx’s proven privacy and verification IP into AI assurance, creating a new growth platform for the company in a rapidly expanding market.