“Your AI agent is running. Do you know what it’s doing?”
AgentPentest is a self-serve software platform for teams shipping autonomous agents. Run standardized evaluations, harden the model, and export the evidence and training data, all from one workspace.
Each module is a standardized part of the platform. Enable what you need. There is no custom scoping call.
Connect an agent endpoint and run a standardized adversarial battery. The platform returns a signed behavior report with replayable evidence.
Turn findings into fixed behavior with a standardized alignment pipeline that runs on your model or a managed one.
Export execution-verified, multi-agent trajectories straight from the platform: standardized, scored and training-ready.
Every workspace runs the same four-step pipeline. Modules can be used on their own or end to end.
Point the platform at an agent endpoint and launch a standardized threat matrix.
Get a signed behavior report with replayable evidence and prioritized findings.
Apply the alignment pipeline to fix the failures the scan surfaced.
Export EU AI Act Annex IV documentation and a signed misuse-risk verdict.
Every run produces artifacts a security team, a procurement board or an auditor can actually work with.
A scored, cryptographically signed report across six behavioral dimensions.
The raw, replayable run records behind every finding: tamper-evident and auditable.
A prioritised remediation list, not just a pass/fail score.
The standardized alignment pipeline applied to the failures the scan found.
Technical documentation structured for Annex IV review, generated from real evidence.
A signed jailbreak and misuse risk verdict for procurement and sign-off.
Transparent, self-serve pricing: evaluate first, scale when you ship.
For a solo engineer or an early-stage team.
Request early accessContinuous evaluation for a shipping team.
Request early accessScale, governance and procurement.
Talk to usAgent adoption is outpacing the tools to verify what agents actually do. Regulation is arriving. AgentPentest sits between the two.
Swarm runs produce execution-verified trajectory data.
We use that data to train stronger, safer agent models.
Those models independently test other companies’ agents for behavioral risk.
Every test run generates more trajectories. The loop tightens each time.
EU AI Act Annex IV documentation and auditable evidence are platform outputs, so compliance becomes a reason customers buy.
A European product for teams that need control over their data, models and supply chain.
The platform is live and self-serve. Create a workspace and run your first evaluation in minutes.
Create a workspace, connect one agent endpoint and run a standardized scan. Upgrade when you need more runs, reports or exports.
