Agent Trust Assurance
Test an agent before production the way any other privileged system is tested, and re-test it when it changes.
What it does
Capabilities
- Security tests for prompt injection, privilege escalation, data leakage and unsafe external calls
- Permission tests that try to exceed the role and cross the tenant boundary
- Tool behaviour tests for dangerous calls the task never required
- Regression testing between agent versions
- Certification from development through internal, approved, production and high-risk
- Recertification when the model, prompt, tools, MCP servers, permissions, data sources or owner change
What it leaves behind
Evidence produced
- A certification record per agent version
- A diff between versions with the resulting risk change
Certification, and what re-opens it
Development
Under construction. No production data, no production tools.
Internal
Cleared for internal use against non-regulated data.
Approved
Passed security, permission and policy testing.
Production Certified
Cleared to run production traffic under monitoring.
High-Risk Certified
Cleared for sensitive actions, with the extra review that implies.
A change to any of these re-opens certification
Specification
Where this is specified
2 blueprint pages carry the specification for this module.
Agent Lifecycle Management
Agents should have controlled birth, change and retirement processes just like employees and service accounts. Lifecycle automation prevents dormant agents, obsolete permissions and orphaned credentials from accumulating as organizations rapidly experiment with AI.
LIFECYCLE PRINCIPLE →
Secure SDLC & Product Security
Customers will evaluate Agent Trust Cloud as critical security infrastructure. Product security must therefore be a company capability, not a checklist added before enterprise sales. Build controls into development, release and operations from the first production version.
TRUST SALES →
13 modules and services make up the platform. See how they fit together.