Outerlimit Raises $16 Million to Stop Rogue AI Agents From Causing Harm
Outerlimit, a new security firm, has raised $16 million to address the growing challenge of rogue AI agents. Their solution focuses on a decentralized security layer that proactively prevents harm by continuously monitoring and enforcing policies on autonomous agentic AI, regardless of whether the agents are aligned or misaligned. They achieve this through a ‘discover, observe, enforce’ system, guaranteeing the scope and conditions of agent actions.
Outerlimit has emerged from stealth with $16 million in pre-seed funding, backed by investors including AlbionVC, Evolution Equity Partners, Crane Venture Partners, and individual angel investors. Founded by Tony Pepper, Neil Larkins, and Peter Vincent, the company is tackling a critical issue: the potential for harm from increasingly powerful and autonomous AI agents.
As enterprises rapidly adopt autonomous agents to accelerate decision-making and automate tasks, these agents are gaining wider access to business identity and credentials, and their complexity is increasing. This complexity makes it difficult to define and limit what actions these agents can take – a process known as ‘alignment.’
Traditional cybersecurity relies on fear of consequences and laws to control human behavior, but autonomous agents operate without a concept of rules or morality and are not deterred by the fear of punishment. Outerlimit’s approach shifts the focus from controlling behavior to preventing harm, regardless of whether an agent is aligned or misaligned.
Their system utilizes a ‘discover, observe, enforce’ framework. It locates agents within a system, monitors their behavior, and enforces a pre-defined policy of allowed and disallowed autonomous actions. Crucially, Outerlimit can guarantee that if an agent is delegated a token, it can guarantee the scope and location under which that token can be used.
This approach sidesteps the alignment problem – the difficulty of ensuring an agent’s goals remain aligned with human intentions – by treating all agents similarly, focusing on preventing harmful actions regardless of the agent's internal state. By operating at machine speed, Outerlimit can proactively block potentially damaging actions before they occur.
The underlying goal is to enable businesses to safely harness the power of AI agents, emphasizing trust as a key factor in a future dominated by co-intelligence. The company believes that success should be measured by an organization’s ability to safely deploy AI agents, not by the sheer number of agents deployed.