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

Humanos raised $3.2 million in seed funding to assess the risk of AI agents in real time. Humanos has raised $3.2 million in seed funding to develop live risk scoring for AI agents. The Lisbon-based startup is targeting organizations that need to understand the behavior of agentic systems and, potentially, incorporate that information into insurance and governance decisions.

Measuring an agent by what it does

Agentic systems can call tools, access data and take actions, creating a larger operational surface than conventional chat interfaces.

The company’s central idea is to assess an agent based on how it operates rather than treating the underlying model as the entire security boundary. That is a useful distinction because the same model can have very different risk depending on which tools it can access, what data it can retrieve and what actions it is permitted to perform.

Humanos’ product information describes the risk-scoring approach and its intended role in assessing autonomous systems.

The company is entering a market where enterprises are looking for ways to measure agent behavior before allowing agents to perform sensitive tasks.

• Humanos has raised $3.2 million in seed funding to develop live risk scoring for AI agents.
• Humanos raised $3.2 million in seed funding to assess the risk of AI agents in real time.
• The Lisbon-based startup is targeting organizations that need to understand the behavior of agentic systems and, potentially, incorporate that information into insurance and governance decisions.

Where risk scoring fits into AI governance

Humanos is building around a problem that becomes more difficult as AI agents gain permission to take actions. A conventional software application has a relatively predictable execution path, while an agent can choose tools, retrieve information and make decisions during a task.

What the risk score is meant to capture

The company’s approach is to score risk based on agent behavior rather than treating the underlying model as the complete security boundary. That can give organizations another signal when deciding whether an agent should be allowed to perform a sensitive operation.

Risk scoring could become a useful input for governance, but a single number cannot replace technical controls. Security teams still need logs, permissions, testing and a way to stop or contain an agent when something goes wrong.

The insurance angle makes the measurement problem especially interesting. Insurers need evidence that can be evaluated consistently if they are going to assess systems that can act autonomously. Humanos is positioning its technology around that emerging requirement, but its usefulness will depend on how well the scores correlate with real operational risk.