Test a configured agent with Fisher, assess a model and its safeguards, or discuss how to reduce risks from your own agents acting together.
Fisher tests configured, tool-using AI agents with adaptive multi-turn attacks across permissions, data boundaries, and workflow rules. It records the conversation and the tool actions it can observe, replays confirmed behavior, and returns findings, replay results, and remediation guidance for security, governance, and release decisions.
We agree access and testing conditions before any testing begins. Pricing is available on request.
Assess a model and its safeguards against agreed requirements and attack conditions. Use the evidence to inform adoption or release decisions. Model-level results do not replace testing the configured agent.
Help prevent harm from your own agents working together. Evaluate unauthorized collaboration, shared-state risks, and containment and recovery controls while preserving legitimate work. Scope is agreed for your multi-agent system.
An agent can leave information that another run uses later. Our counter-swarm approach examines that related activity, the task rules governing its use, and the state that survives a restart.
The goal: contain unauthorized collaboration while preserving useful work.
Deep Model Trust is mace AI’s technical architecture for moving from testing agent behavior to building systems whose authority, decisions, and release processes are bounded and verifiable. Fisher is the first product built around that thesis.
Fisher’s adversarial evidence informs how we’re exploring a trustworthy agent architecture.
Tell us about the model, agent workflow, or multi-agent system and the behavior you need to test.
Book demo →