THE CRUNCH
New York startup Outerlimit has emerged from stealth with $16 million in pre-seed funding to build a decentralized security layer for autonomous AI agents. The startup aims to prevent harmful actions by discovering agents, observing their behaviour and enforcing pre-defined policies at the moment of execution. Founders Tony Pepper, Neil Larkins and Peter Vincent argue that traditional security based on human rules or
fear of consequences does not work for AI agents, which operate at machine speed and can change behaviour dynamically. Outerlimit treats all agents as tools that must be bound to a specific policy, sidestepping the difficulty of aligning their internal coding. The funding was led by AlbionVC, Evolution Equity Partners and Crane Venture Partners.
The startup's approach focuses on guaranteeing the scope and conditions under which an agent can use delegated credentials. By operating at the same speed as the agent, Outerlimit claims to be one step ahead of potential misuse. The founders argue that the next phase of enterprise AI should be measured by the ability to safely harness agents rather than the number deployed.