Outerlimit raises USD $16 million to secure AI agents

Outerlimit has emerged from stealth with USD $16 million in pre-seed funding, which it described as among the largest pre-seed rounds in cybersecurity.

The London-based start-up focuses on security controls for autonomous AI agents, a fast-growing area as companies move from testing AI systems to connecting them to internal tools, data and application programming interfaces. Outerlimit was founded by Tony Pepper and Neil Larkins, who previously led Egress Software, and Dr Peter Vincent, a theoretical neuroscientist.

The round was backed by AlbionVC, Evolution Equity Partners and Crane Venture Partners. Outerlimit is building a decentralised security and authorisation layer designed to control what AI agents are allowed to do when they call a tool or API.

Its pitch centres on what it sees as a gap in current enterprise security models. As businesses give AI agents access to systems and workflows, existing identity, access and governance tools may not provide enough visibility or control over actions taken automatically and at speed.

Outerlimit said its design applies Zero Trust principles at the point of execution. Instead of storing credentials, keys or secrets in one place, it fragments them across the wider agent environment and reconstructs them only when a tool is invoked and the required conditions are verified.

That ties authorisation to a specific action rather than only to a user or system identity. The company argues this could give organisations a way to approve or block agent actions in real time as they move from experimentation to operational use.

Dr Peter Vincent, Founder and Chief Technology Officer, outlined the case for a different approach to security in AI systems.

"We've spent decades building security systems around human qualities such as loyalty, trust, compassion, and even fear which drive predictable decision-making," said Dr Peter Vincent, Founder and Chief Technology Officer, Outerlimit. "In this new agentic era, where agents use reasoning models with access to tools that can impact the world, they simply don't operate within these human constructs. Agents can change their behaviour based on what they read, or how they interact with other agents, while acting at machine speed. Harnessing this powerful intelligence requires a fundamentally new security architecture - one that binds identity, authorisation, and action into a single operation at the moment of execution."

Control path

Outerlimit said its product is designed to give customers a staged route to tighter controls. It described three steps: discovering agents, tools and shadow AI; observing activity and preserving chain integrity across multiple steps; and then enforcing controls on each agent action.

The emergence of vendors focused on agent security reflects a broader shift in enterprise AI. Many companies now want AI systems not only to answer questions or generate content, but also to take actions such as querying databases, using software tools, updating records and carrying out routine operational tasks.

Those use cases increase the potential value of AI, but they also raise the stakes for security and compliance teams. If an agent can act across systems, errors or misuse can have direct operational consequences, pushing access controls and runtime oversight higher up the agenda.

Tony Pepper, Founder and Chief Executive Officer, said many organisations are already under pressure to permit wider AI use while lacking confidence in current safeguards.

"The demand to deploy agentic AI is unparalleled, but security teams are being asked to sign off on risks they cannot adequately control," said Tony Pepper, Founder and Chief Executive Officer, Outerlimit. "Blocking adoption isn't a strategy, it simply drives the use of unsanctioned AI outside of enterprise oversight. Today's launch of Outerlimit represents a paradigm shift to securing agentic AI, creating a world where AI agents can be trusted implicitly because every action is provably observed, controlled, and compliant without limiting their expressiveness."

Backers and founders

AlbionVC Partner Ed Lascelles said investors were backing both the founding team and the scale of the market opening around agentic AI security. Pepper and Larkins previously built Egress Software, which was acquired by KnowBe4 in 2024. Vincent's background is in theoretical neuroscience, with a PhD linked to the Sainsbury Wellcome Centre and the Gatsby Computational Neuroscience Unit at UCL.

The company also disclosed support from a group of strategic angel investors, including figures from Alan, Mistral, ReliaQuest, A-Lign, Goldman Sachs, FTV Capital, Twin Track Ventures, Garrison and GP Bullhound.

Investor interest in the sector has risen as enterprises try to reconcile the promise of more autonomous software with the limits of existing governance models. Outerlimit is positioning itself around the view that the next stage of AI adoption will depend less on building new agents than on proving their actions can be monitored and restricted.

"The next phase of enterprise AI should not be a race to build the most agents. Instead, success should be measured by the ability of an organisation to safely harness the full power of their agent deployments," said Dr Vincent. "If intelligence is to become commoditised, trust will be the limiting factor. As we accelerate into a new era of co-intelligence, getting this right is a fundamental obligation for the sake of individuals, organisations, and international security."