The business case for a security product is a difference: the risk register for an agent deployment without it, and the register with it, from the operator who is paged to the board that has to defend it. This section writes that difference down, computed from a public model, for our own product first and then for others’, in their own words.
Not a feature list: which entries leave your register, whose they were, and what reaches the board. And which ones a product only makes less likely, said as that.
Many security products are sold on capability. This maps the capability to the risks it changes, at every altitude, in terms a CFO and a board can read. If it is wrong about you, it changes.
No case can be computed until somebody has answered the questions truthfully about one agent in one deployment. That answer is what an Agent Behaviour Policy produces.
The model is the RiskGraph Explorer’s, from one of our live demos, copied into this site with its provenance. It is small on purpose: sixteen questions, readable in one sitting, so an argument about a case is an argument about an answer, not about a formula. Nothing on these pages is a score, and no case is a statement that a product works; it is a statement of what changes in the register if it does what its documentation says.
Our own product first, so the method is tested on us. Then open-source projects, OWASP’s first, which anybody can deploy and nobody has to pay for, but which cost something to adopt and more to customise; each case says what. Cases about commercial products are drafted from their own documentation and sent to the company before they are listed.
Our own
Open source
Across the open-source projects, the answers that move are egress, access to data, the record, the account, stopping and undoing. None of them moves who owns the stop, the side effects of stopping, the procedure after it, or the class of data in reach. Those are decisions and documents, not software, which is where a behaviour policy and a licence to operate come in. How OWASP’s own projects relate to each other, and to these answers, is mapped in OWASP, as a graph.
RiskMandate publishes its behaviour policies and toolkit openly and sells the work on top. These companies run the same model around projects in this section, or beside OWASP. They are here because the conversations the lead wants are with them: a case about their project is also a case for what they sell.
The patterns, as facts. Some stay inside OWASP with a company alongside: DefectDojo, SAMM with Codific, secureCodeBox with iteratec. ZAP left OWASP in 2023, saying it could only be in one foundation, and became ZAP by Checkmarx in 2024. Several gave their project to a foundation and sell a distribution of it: Falco, Kyverno, agentgateway. Some sell a hosted control plane over an open engine. Two such companies closed in 2025: after the creators of Open Policy Agent joined Apple, OPA stayed a CNCF project with no change to its governance or licence; Aserto wound down as a commercial entity.
Twelve kinds of product that could sit around an agent, each computed against the typical deployment with its own questions answered as they would be without it. Every row is our reading of what the category does, not any vendor’s claim, and the change is conservative on purpose: where a category could move an answer to partly or fully, the row says partly.
Two patterns are worth reading off the table. Most categories retire one or two entries each, so the case for any single product is narrow and exact rather than broad, and a register gets small by combining them. And every category adds something: a service in the request path, a store of prompts, a credential with wide reach. The case is only honest with that column in it.
If you build a product for agent security and would like its case written, or you have read a case about your product and it is wrong, write to us. If you run agents, the case for anything starts with a map of one of them.