Exabeam Inc. is asking a blunt question about artificial intelligence agents before they ever touch a live system: does the thing actually do what it was hired to do? The company calls the discipline Agent Behavior Verification, or ABV. And on Monday it released the open-source tool that makes the question answerable — Praxen.
Praxen is not another runtime monitor. It runs before deployment.
That is the key distinction Exabeam is drawing. The company already sells Agent Behavior Analytics, a product that watches agents after they go live. ABV is the pre-deployment check.
Praxen is the reference build of that check, released under the Apache 2.0 license. The tool starts from what Exabeam calls an ABV remit.
Think of it as a policy contract. It sets out what an agent may do, what it may touch, and where it has to stop. Praxen then measures that remit against the agent’s real tools, configurations, memory, integrations, and operating environment.
Where the two diverge, the tool reports specific findings, recommends fixes, and assigns a maturity score for the agent’s security posture. This is a different question than runtime monitoring answers. Runtime tools ask: is the agent behaving badly right now?
ABV asks: was it ever capable of behaving well in the first place? The distinction matters because an agent can pass every live check and still be structurally misaligned with its job description.
Praxen tries to catch that gap before anyone hits deploy. The timing is notable. Exabeam has been pushing hard through 2026 on agent security.
In April it expanded its Agent Behavior Analytics product to cover agents across ChatGPT, Microsoft Corp.’s Copilot, and Google LLC’s Gemini. Weeks later it stretched that coverage across Google Cloud’s entire agent ecosystem.
Now comes the pre-deployment layer. Praxen is an open-source reference build. That means anyone can inspect the code, modify it, or build on top of it.
Apache 2.0 license terms are permissive. The company is effectively publishing its methodology for public scrutiny and use. That is a different strategy than keeping the tool proprietary and selling it as a product.
The report does not name the Praxen developers. It does not give specific numbers on how many agents the tool has tested internally.
It does not offer customer quotes. What it does offer is a clear line: Exabeam believes the industry has been skipping a necessary step. Deployment security is not enough if the agent was broken before it ever started.
Whether Praxen gains traction will depend on whether other companies adopt the ABV remit structure. A policy contract is only useful if someone writes one.
Exabeam has provided the tool to write the contract and the tool to check the contract. It has not provided the contract itself. That part falls to each organization deploying an agent.
The maturity score is interesting. It is a single number that summarizes how well the agent’s real capabilities match its assigned role. Low score means the agent can do things it was not hired to do.
High score means the remit and the reality are aligned. The report does not specify what constitutes a passing score or how the scoring algorithm works.
That detail is presumably in the code. Exabeam is betting that the market will care about pre-deployment verification as much as it cares about runtime monitoring. The two are not substitutes.
They are sequential. First verify.
Then monitor. Praxen is the verification half, now available for anyone to run.
