Embed
Headless runtime, executor integration SDK and standalone verifier. Bind action, authority, current state and outcome.
OntoGuard · Enterprise and OEM infrastructure
Your customers will ask whether a legitimate agent can misuse a privileged tool. Add the control boundary that checks the exact action before effect.
Headless runtime, executor integration SDK and standalone verifier. Bind action, authority, current state and outcome.
Your customer, product UI, policy authority, execution route, monitoring and deployment ownership.
Typed action schema, trusted-context adapters, executor verification, TRACE evidence and OpenShell supervisor integration.
Decision Authorization includes an OpenShell-compatible supervisor adapter and an agent-action integration. It is an integration capability; it does not imply an NVIDIA partnership, listing or certification.
Start with one protected action. Exercise exact authorization, changed parameters, stale authority and outcome evidence. Measure latency and integration work in your environment.
Embedded OEM or field-of-use licensing is negotiated privately. No public pricing.
See reference measurements. Measure added overhead at your target; no production SLA is implied.
Run inside your controlled environment; you own the protected routes and trusted sources.
Support, deployment obligations and escrow terms are subjects for the licensing agreement. No escrow commitment is made here.
| Technique | What the attacker gets | What OntoGuard does | Limit |
|---|---|---|---|
| AI Agent Tool Invocation (AML.T0053) | A credentialed agent calls a privileged tool. | Check the exact requested operation, target and current authority; refuse or escalate. | Only protected execution routes are checked. |
| Valid Accounts (AML.T0012) | Legitimate account access is used for an unauthorized change. | Separate authentication and tool access from permission for the exact action. | Not credential theft prevention. |
| LLM Prompt Injection (AML.T0051) | Injected instructions propose a changed payment or record. | Check proposed parameters against trusted authority rather than prompt claims. | Not universal prompt injection detection. |
ATLAS mappings describe where OntoGuard can act; they are not a certification or a guarantee of detection.
Mapping data · MITRE ATLAS · Illustrative protected-action examples.
Public evidence covers a reviewed integration and reference tests, not a customer deployment. See evidence scope.