Humanos Raises $3.2 Million to Build Live Risk Scores for AI Agents

An AI agent that behaved safely yesterday can behave unpredictably today, a problem that makes a one-time security review effectively meaningless for software that keeps changing its own behavior after deployment.

Lisbon-based Humanos closed a $3.2 million seed round led by Anthemis, with Springbank, Critical Ventures and Start Ventures participating, alongside angel backing from founders of Sword Health, Feedzai and OutSystems. The company assigns each AI agent a live risk score between 0 and 100 that shifts continuously as the agent operates, weighing who is behind the agent, what permissions it holds, its past behavior, and whether its most recent activity looks unusual relative to its own history.

Why a Static Review Cannot Solve This Problem

Traditional security review evaluates software at a fixed point in time, a model that assumes a system's behavior on the day of review will hold going forward. An AI agent breaks that assumption directly: it can be reconfigured, granted new permissions, or simply drift in how it behaves as it processes new data and situations it was not explicitly tested against, all without triggering the kind of code change a traditional security review process is built to catch. Humanos's live, continuously updating score is a direct response to that gap, treating an agent's risk profile as something that needs ongoing measurement rather than a one-time certification.

What a "Risk Passport" Actually Captures

The company's risk passport concept builds a running history of the controls placed on a given agent, any incidents linked to its activity, and eventually the economic outcomes its actions produced, effectively creating a track record an enterprise can reference before granting an agent expanded permissions or before onboarding a new agent from an unfamiliar vendor. That framing treats agent risk the way a credit bureau treats borrower risk: not as a single yes-or-no judgment, but as an evolving record that updates as new behavior accumulates, letting a business make a more informed decision about how much autonomy to extend to a specific agent based on what it has actually done rather than what it claimed it would do at launch.

What This Means for Founders

Humanos's raise is a useful signal for any founder building infrastructure to govern a category of software that changes its own behavior after deployment, a problem traditional security and compliance tooling was never designed to solve. As more companies deploy AI agents directly into production systems, the gap between a one-time review and the ongoing reality of how an agent actually behaves is exactly the kind of structural gap that tends to produce durable infrastructure businesses, since the underlying need does not go away once the first version of the product ships. A founder building governance, monitoring or trust infrastructure for autonomous software should study Humanos's decision to frame the problem as continuous measurement rather than point-in-time certification, a framing that is likely to apply well beyond AI agents to any software category where behavior meaningfully changes after deployment.

Founders building AI infrastructure, security, or enterprise trust and safety tooling looking for investors with an active thesis in the space can browse cybersecurity investors on AngelLinx, or explore active capital on AngelLinx @ angellinx.ai/register.

AngelLinx Intelligence | angellinx.ai