Agentic AI Security & Runtime Review
Review what the system can do, not only what it can say.
DDAI assesses the action surface of tool-using and agentic systems: identities, permissions, workflow states, external effects, human authority, failure routes, and the evidence needed to reconstruct a consequential action.
The problem
The risk changes when the system can alter external state.
A text response may be wrong. A tool-using system can send a message, change a record, trigger a payment, modify code, approve a workflow, or pass authority to another component. The control question is therefore not only whether the model produced a good answer. It is whether the system had permission, whether it had authority, whether a human should have approved the action, and whether independent evidence exists.
Review areas
Intended purpose and action inventory
Non-human identities
Tool and data access
Least-privilege permissions
Planning and execution separation
Workflow states and retry behaviour
Multi-agent delegation
Prompt injection and indirect instruction risk
Sensitive-data exposure
Human approval and override
Monetary, rate, and impact limits
External execution receipts
Monitoring, incident, and kill-switch design
Runtime policy and decision evidence
Change and drift triggers
Delivery sequence
1. Scope and written authorisation
2. Architecture, action, and evidence review
3. Controlled defensive testing where agreed
4. Findings, severity, and business impact
5. Remediation and retest
6. Runtime-evidence and Evidary roadmap
Deliverables
Action and tool inventory
Identity, permission, and authority map
Threat and failure model
Human-oversight gap analysis
Runtime policy and evidence requirements
Findings and remediation report
Retest record where contracted
Procurement or governance summary
Evidary-ready runtime evidence design
Evidary Governance Harness boundary
Where a recurring evidence workflow is justified, transition planning can be scoped separately with Evidary. The review remains provider-neutral, and no evaluator replaces accountable human judgement.
Govern the action chain, not only the model.
DDAI can help you connect business value, technical implementation, human oversight, and governance evidence from day one.
Review a tool-using system