SECURITY ATLAS / EVALUATION BRIEF
Noma AI Agent Security Platform
Research reviewed: 2026-10-18
https://atlasofsecurity.com/landscape/vendors/noma-agent-security/

THE ATLAS TAKE
Noma is a useful evaluation candidate when teams need an inventory and AI security posture view across a defined agent environment, with possible extension into runtime controls. Its public material separates AI-SPM, Agent Access Control and Open Enforcement integration paths, which should shape the architecture review. Atlas would first establish what the selected connectors can actually discover and which permissions substantiate a finding. Enforcement through hooks, gateways, SDKs or APIs then needs its own test. Platform breadth is promising, but does not prove every agent version and route is covered.

SCOPE
Assess discovery and AI-SPM for a defined environment. Open Enforcement integrations and AI-DR are complementary modules whose coverage must be verified separately.

WORTH A LOOK WHEN
Teams mapping AI assets and posture with willingness to engineer explicitly selected enforcement integrations.

THINK TWICE WHEN
The plan treats a discovered asset or posture finding as proof of blocked runtime behavior.

IMPLEMENTATION
1. Inventory a bounded environment
List known agents, versions, tools and identities and compare them with the connector output.

2. Validate one risk precisely
Tie the finding to an actual permission or configuration and a resource the agent can reach.

3. Choose and test enforcement
Record the selected integration point, failure behavior, target-side result and reversal procedure.

OPERATING REALITY
Platform owners maintain connectors and inventory freshness while application owners validate findings and control points. Agent versions, tool registrations and identity permissions change independently, so a meaningful case needs the affected version and route. Each chosen enforcement integration also needs monitoring, error handling and a support owner.

COMMERCIAL CONSIDERATIONS
Obtain a module and integration-specific quote. Include deployment responsibilities for the chosen Open Enforcement path and separate posture, access-control and detection/response scope.
- Connected AI estate and selected modules
- Chosen enforcement integration and deployment
- Connector upkeep, finding validation and response operations

EVALUATION PLAN / PROPOSED TESTS, NOT COMPLETED RESULTS
1. Discover a known agent and its tool relationship
Acceptance evidence: Record version and authoritative configuration evidence.
Owner: __________  Pass criterion: __________
Observed result: __________  Evidence location: __________

2. Reduce a safe excessive test permission
Acceptance evidence: Verify effective resource access and updated finding state.
Owner: __________  Pass criterion: __________
Observed result: __________  Evidence location: __________

3. Exercise one selected enforcement route
Acceptance evidence: Capture policy, failure handling and the independent target effect.
Owner: __________  Pass criterion: __________
Observed result: __________  Evidence location: __________

QUESTIONS FOR THE VENDOR
- Which assets are found without application changes?
- Which control points enforce this policy?
- How are session findings tied to the affected agent version?

EVIDENCE GAPS
- Atlas has not independently tested discovery completeness or runtime control efficacy.
- Confirm supported connectors, integration modes and release-specific module entitlement.

SOURCES
Noma: platform and Open Enforcement (Product information)
https://noma.security/
Noma: AI-SPM asset and risk model (Product information)
https://noma.security/products/ai-spm
Noma: Agent Access Control registry and policy (Product information)
https://noma.security/products/agent-access-control

Use synthetic fixtures in an authorized environment. Agree rollback and acceptance criteria before enforcement.
Documentation research and Atlas analysis. Independent hands-on effectiveness, effort, and complete price remain unassessed.