Verifoxx Delivers Sovereign AI Proof of Concept, Advancing Runtime Assurance for AI in Constrained Environments

3 April, London, UK – Verifoxx today announced the successful completion of its SentinelOps Proof of Concept (PoC), developed under the UK Government’s Sovereign AI – Proof of Concept programme.

Selected as one of 25 projects from over 1,100 applicants, SentinelOps focuses on a core challenge in modern AI systems: how to verify behaviour at runtime in environments where telemetry, connectivity, or trust cannot be assumed.

SentinelOps introduces a runtime behavioural assurance capability for AI workloads operating in constrained and security sensitive environments. It enables organisations to determine whether systems are behaving as expected at execution time, without relying on logs, telemetry, or centralised infrastructure.

The PoC delivered four core capabilities: runtime observation, behavioural modelling, deviation detection, and policy-constrained outputs; enabling verification of system behaviour without exposing raw data.

The capability establishes a new assurance layer focused not just on system identity or integrity, but on whether runtime behaviour remains within an admissible policy and operational envelope over time.

SentinelOps builds on Verifoxx’s existing work in privacy enhancing technologies and zero-trust systems, extending its approach to enabling trust into the AI domain. It is designed for environments where systems, operators, or control planes must decide whether to rely on machine-generated outputs under conditions of limited visibility.

Bala Subramanyan, CTO at Verifoxx, said: "The challenge is no longer about collecting more telemetry, but about establishing whether a system’s output can be trusted in the first place. SentinelOps is designed as a runtime assurance layer for environments where other systems, operators, or control planes need to decide whether to rely on machine-generated outputs under conditions of limited visibility. It builds directly on our work in zero-trust systems, policy-driven infrastructure, and privacy-preserving compute."

With the PoC complete, Verifoxx is now engaging with industry partners to validate SentinelOps in real-world environments, particularly across industrial edge, distributed systems, and other constrained execution contexts. These early collaborations are intended to test the capability under representative operational conditions and generate evidence of practical applicability.

Alongside this, Verifoxx is exploring follow-on funding opportunities to progress SentinelOps toward demonstrator and production readiness. This next phase will focus on integrating the capability into live environments, refining its application across different system architectures, and establishing its role as part of a broader runtime trust infrastructure for AI-driven systems.

To read more about the Innovate UK competition that funded SentinelOps, click here.

Register your interest as a SentinelOps Industry Partner