Mace Ai · Model Assurance on Cloud and Edge

Your models and agents,
tested and verified.

Advancing trustworthy AI through adversarial testing and remediation.

Our mission

Mace Ai helps enterprises deploy AI models and agent harnesses safely. Our mission is Deep Model Trust: building systems and evidence for AI agents operating with real authority across tools, memory, shared state, changing models, and remediation cycles.

Built by leaders from world-class companies and universities.

Meet the team →
our adversarial red-teaming product

Fisher

Vulnerability discovery and remediation of agentic surfaces for highly regulated industries.

Fisher tests configured, tool-using AI agents through adaptive multi-turn conversations. It examines traces and tool calls, uses evolutionary search to discover vulnerabilities, and turns the results into findings, validation evidence, and remediation guidance.

Learn moreRead the white paper →
Fisher interface example - findingsExample: Status of individual findings from a HIPAA Test assessment.
Three Fisher findings from the HIPAA Test assessment, with Observed, Confirmed, and Exploration signal evidence labels and separate replay statuses.
01
Test real agent workflows

Adaptive multi-turn conversations probe the agent, its tools, and its operating boundaries.

02
Discover vulnerabilities

Traces and tool calls are examined through evolutionary search to find novel attack paths.

03
Remediate with evidence

Findings include replay evidence and guidance your team can use to fix and retest.

Our Research

Our work on adversarial testing, model security, and autonomous systems.

Published work
Fisher white paper
Securing Tool-Using AI Agents
The Fisher testing method: adaptive multi-turn testing, action-level evidence, and remediation re-attack.
Open white paper →
Research paper
Counter-Swarm Doctrine:Containing Coordinated Agent Intrusions
September 5, 2026
Greg Frank proposes an approach to discovering and containing unauthorized coordination across AI agent executions.
Open research paper →
R&D pipeline
Private Model
Private model systems for regulated environments.
An efficient and portable small model with installable domain specialist data packs in financial services, law, healthcare, software engineering, and more. Predictable, private, and abstaining from action when critical input is missing. Designed for scoped, domain-specific deployments where teams need useful reasoning without handing broad authority to a general-purpose model.
Open roadmap →

All research and articles →

Get started

Your first warning should not come from a client.

Start with the model, agent workflow, or agent behavior that concerns you.

Contact Us →