Overmind co-founders Sam Brunt, Akhat Rakishev and Tyler Edwards
As autonomous AI agents begin to move from experiments into real-world deployment, a growing problem is emerging: once these systems are live, it’s increasingly difficult to understand what they’re doing, why they’re doing it and whether they can be trusted to behave safely in high-stakes environments.
It’s not just personal assistants like OpenClaw (/Clawdbot/Moltbot), agents are rolling out across verticals, and in highly regulated industries like legal, defence, healthtech and fintech, this can pose operational and security risks.
This is the space where London-based Overmind positions itself, providing a “supervision layer” for agentic AI.
The startup, which launched just seven months ago, raised a £2m seed round led by specialist cybersecurity investors, Osney Capital, with participation from 14Peaks, Portfolio Ventures, Antler and Endurance Ventures.
Overmind, self-described as “supervision for super intelligence”, is taking a different approach to monitoring agentic AI. While traditional approaches to AI security were built for deterministic software, agentic systems are inherently non-deterministic by design — they adapt, learn from new data, update their memory and change behaviour constantly — which, according to co-founder and CEO Tyler Edwards, requires new, adaptable layers of protection.
“It's about moving security from these kind of static fire-and-forget tools to something that is constantly running, constantly learning, constantly understanding,” Edwards tells Pathfounders. “Because this is software that isn't static. It changes, it adapts, it learns new contexts, updates its memory. It's not the same as you found it yesterday.”
Because there are so many potential risks, it is impossible to prepare an agent for all of them, but it is possible to assess what it is doing when it’s in production, Edwards explains. Just like with a human, while you can’t prepare for everything that might possibly go wrong, you can put fail safes in places to prevent critical errors.
Edwards, who spent eight years building AI systems for British intelligence agencies including MI5, MI6 and GCHQ, has co-founded Overmind alongside CTO Akhat Rakishev and CRO Sam Brunt.
Rakishev previously led machine learning infrastructure at Monzo and Lyst, while Brunt has scaled go-to-market at unicorns Funding Circle, Pipe and Vertice.
Having traded in the stress of building AI systems for British intelligence agencies — “You come in one day and they say, we've got intelligence that within the next week Russia is going to invade Ukraine. Here's what we need to do before that happens” — for, as Edwards puts it, “a slightly different flavour” of stress in the form of a startup, his background in defence has shaped the company’s focus on solving problems for complex, high-stakes industries, but without the constraints of civil service.
Entering a rapidly emerging market, companies like AgentOps.ai (the creators Agency AI raised $2.6m in 2024) are targeting agent observability and testing, while Zenity, an agent-focused security startup, raised $38m last year.
Overmind specialises in AI models for specific tasks and sectors, enabling fine-tuning down to the individual company level, dependent on the data available. The startup plays in a space where security is key and generalist models might not cut it.
“Say for example, you're building in medicine,” Edwards says. “Your model doesn't need to know the full works of Shakespeare. Maybe it just needs to know specifically what a melanoma is and how to diagnose it.”
The AI can learn the specific tasks and teams can ship smaller, more specialised models that are cheaper, faster and easier for the teams to build themselves.
“In the new frontier of autonomous AI, agent security, performance, and execution are the ultimate competitive advantages,” Osney Capital partner Adam Cragg said in a statement. “Overmind provides businesses with truly differentiated technology that monitors and secures agentic AI while iteratively improving model performance, enabling teams to scale with confidence.”
The developer-focused software development kit (SDK), integrates into existing codebases and is designed to enable AI and agents to learn from production data — it can learn from its surroundings and specialise in specific tasks.
“We kind of just want to get out of the way,” Edwards says. “Here's the tech. Go crazy and we'll get out of the way. But here are the tools for you to just do really incredible things.”
The team is also building a telemetry stack to improve reinforcement learning by collecting more and higher quality data and the fundraise will go towards technical hires for further product development.
Company information:
Six staff, with plans to double headcount by the end of the year
Based in London