AI agent security
Trust boundaries, least-privilege tool permissions and human approval points for autonomous workflows, reviewed against the irreversible actions an agent can reach.
Review AI agents- Standards
- ISO/IEC 42001 · NIST AI RMF
- Testing
- Adversarial, tool-aware
Why this matters now
AI is already in use inside most organisations, usually before governance catches up. AI agent security gives you control without blocking work that is genuinely useful.
What the engagement covers
Each area below is a defined part of the engagement, with an owner, an output and a date.
Trust boundaries
Where agent input stops being trusted and starts being validated.
Permissions
Least-privilege scoping of every tool and credential an agent can reach.
Data access
Controls on what an agent can read, write and export.
Tool use
Review of tool definitions, side effects and irreversible actions.
Monitoring
Continuous observation with alerting tuned to your risk profile.
Human approval
Checkpoints where a person must confirm before the action executes.
Agent trust boundaries
How the work runs
The same sequence on every engagement, so you know what happens next.
Inventory
Establish what AI is actually in use, including tools adopted without approval.
Assess
Evaluate each use case for security, privacy, legal and model risk.
Control
Apply proportionate controls: policy, access, logging, approval points and monitoring.
Test
Adversarial testing against the deployed system, including indirect injection paths.
Govern
Reporting, review and change control so governance keeps up with adoption.
What you get out of it
Adoption without blind spots
Useful AI work continues; the uncontrolled parts get controlled.
Standards alignment
Mapped to ISO/IEC 42001 and ISO/IEC 27001 so one programme serves both.
Tested, not assumed
Controls verified by adversarial testing against the live system.
Before you enquire
How quickly can this start?
Who actually does the work?
How is this priced?
Where to go next
Review AI agents
One scoping call with the consultant who would run the work. No obligation, no charge.