Overmind raises £2m to supervise AI agents once they go live

On 11 February Overmind, a London-based company that builds software to supervise AI agents, announced a £2m seed round led by Osney Capital, a specialist cybersecurity investor. Also taking part were 14Peaks, Portfolio Ventures, Antler and Endurance Ventures, as Tech.eu reported.

The platform watches how agents behave once they are running in live systems, flags when they deviate and lets teams intervene. It also uses reinforcement learning to improve agent performance and reliability over time, according to the launch coverage.

Tyler Edwards, Overmind’s co-founder and chief executive, put the pitch plainly in the funding announcement carried by Tech Funding News: “The AI security industry is trying to secure the wrong thing.” His argument, in the same statement, is that models will always be open to adversarial inputs, and that most teams have no view of an agent in production whose behavior starts to drift.

Overmind is betting that spending on agent security will follow what agents do in production rather than what goes into the model. The round is small, the company is seven months old and no customer has been named. But the route it has taken since its August 2025 website shows where its founders think buyers’ worries about agents now sit: in live operation, after testing has finished.

From filtering inputs to supervision

Overmind Technology Ltd was incorporated at Companies House on 16 July 2025. By late August 2025 its website described the product as “the security layer for AI”. The pitch then centered on input security, a promise that sensitive data never reaches third-party models and observability across agents’ tool use and reasoning. The company claimed at the time that Overmind “increases protection from malicious inputs by 85%”, without setting out how that was measured.

A blog post on the company’s site, archived in October 2025, set out the idea now at the center of the product. It argued that autonomous agents need “a real-time supervisor” able to tell when an agent is going off track, and that testing a few thousand scenarios before deployment says little about how an agent behaves in live use. The post said financial services, healthcare and other high-stakes sectors were being kept away from agentic AI because existing security approaches could not give them the assurances they need.

By the February launch, that argument had become the product. Input filtering is still part of the offer: the company’s terms of service, last updated in November 2025, list input protection and threat detection alongside agent monitoring, policy enforcement and controls for Model Context Protocol (MCP) gateways. The lead message has moved to supervision of agents in live operation.

The founders and the backers

Edwards founded the company with Akhat Rakishev, chief technology officer, and Sam Brunt, chief revenue officer. Edwards spent eight years building AI systems before Overmind. Rakishev previously built machine-learning infrastructure at Monzo and Lyst, and Brunt helped scale Funding Circle, Pipe and Vertice, according to the launch announcement.

Overmind had six staff at launch and plans to double headcount by the end of the year, as first reported by Pathfounders. The company says the money will go on expanding its technical teams, product development and go-to-market work.

The investors framed the round in commercial terms. Adam Cragg, a partner at Osney Capital, said in the announcement that agent security, performance and execution would be the main competitive advantages in autonomous AI. Adam French, a partner at Antler, said the security and supervision of autonomous agents was one of the most critical bottlenecks in the advancement of AI.

Regulated sectors first

Overmind’s early go-to-market focus is legal, healthcare and fintech. The company’s October 2025 post drew its examples from the same ground: a legal agent quietly weakening liability protections across contracts, a healthcare agent making biased diagnostic suggestions and a financial agent shifting portfolio allocations.

In June 2025 Gartner predicted that more than 40% of agentic AI projects would be canceled within two and a half years, because of escalating costs, unclear business value or inadequate risk controls. A Gartner poll of 3,412 webinar attendees in January 2025 found that 19% said their organization had made significant investments in agentic AI and 42% conservative ones. Gartner also estimated that only about 130 of the thousands of vendors selling agentic AI are real.

Overmind’s product is aimed at the third of Gartner’s reasons, risk controls. Its reinforcement learning feature, which the company says improves agent performance and accuracy, bears on the second. None of the launch coverage names a customer, and the company has not published usage or revenue figures.

On 11 February Overmind had £2m, six people and a pitch it had already rewritten once, from filtering what goes into models to watching what agents do. Its backers, led by a cybersecurity specialist, are paying for the view that oversight of live agents becomes a line in the security budget of regulated firms. Until one of those firms is named as a customer, the £2m is a bet on the founders rather than on proven demand.