Summary
- Outerlimit has raised $16 million in pre-seed funding from AlbionVC, Evolution Equity Partners, and Crane Venture Partners.
- Its architecture aims to bind identity, policy, and execution context at the point where an AI agent calls a tool.
- The approach reflects a wider security problem: conventional identity controls govern access, while autonomous agents can make consequential decisions after access has been granted.
UK-founded Outerlimit has emerged from stealth with $16 million in pre-seed funding to develop security controls aimed at the point where autonomous AI agents invoke tools and take actions.
The round was backed by AlbionVC, Evolution Equity Partners, and Crane Venture Partners. Outerlimit was founded by former Egress founders Tony Pepper and Neil Larkins alongside Dr Peter Vincent, a theoretical neuroscientist who becomes chief technology officer.
The company is attempting to address an increasingly specific security problem. Traditional identity and access management can establish who or what has permission to reach a resource, but an autonomous agent may make many subsequent decisions about how that permission is used.
Outerlimit says its architecture binds identity, authorisation, and action at the moment of tool execution. Rather than relying solely on prompts or the agent’s reasoning process, enforcement is intended to sit in the execution path and make a deterministic allow-or-deny decision when an agent attempts an action.
The company describes the approach as extending zero trust to an “agent action layer”. Its platform is organised around discovery, observation, and enforcement, including identifying agents and Model Context Protocol servers, observing tool calls, and applying controls to actions.
Its cryptographic design also attempts to avoid storing complete credentials in one location. Outerlimit says secrets can be fragmented across the environment and reconstructed only when the required identity, policy, and execution context have been verified.
Those are product and architectural claims from a newly launched company, not independently established performance results. The larger issue they address is nevertheless becoming visible across enterprise AI deployments.
Cyber Insider reported in July that the growth of AI agents and other non-human identities was already putting pressure on identity governance. Autonomous software complicates that problem because an identity may be valid and its permissions technically legitimate while an individual action still falls outside the business purpose for which access was granted.
An agent authorised to query customer records, for example, may still need limits governing which records it can reach, which tools it can call, whether it can export information, and whether it can trigger another agent. Static entitlements do not always express those contextual restrictions.
That creates a control gap between authentication and execution. Several security vendors are now converging on runtime authorisation, tool-call inspection, agent inventories, and traceable delegation as possible answers, although the market remains early and terminology is far from standardised.
Outerlimit’s founders have previous experience building Egress, which was acquired by KnowBe4 in 2024. The scale of the pre-seed round gives the new company substantial early funding, but enterprise adoption and technical effectiveness remain to be demonstrated.
The company’s significance therefore lies less in another AI security launch than in where it places the enforcement boundary. As agents gain authority to query databases, write code, modify cloud resources, call APIs, and initiate transactions, the security decision increasingly has to account for what software is doing now — not only whether it authenticated successfully at the beginning of a session.
