AI Agent Passport Template

What the AI Agent Passport captures

Every autonomous or tool-using agent gets one formal record before it touches your systems. The passport answers the ten questions every AI agent must be able to answer:

• Unique, traceable identity — never a shared human or admin account
• The accountable human owner (business and technical)
• The approved, documented purpose and out-of-scope uses
• Which systems, mailboxes, APIs and data repositories it can reach
• What data it can read versus modify
• Which decisions it can take autonomously, and where human approval is required
• Which credentials it uses, and how they are stored, scoped and rotated
• Whether every action is logged with full attribution
• Where the kill switch is — and whether it has been tested
• When the identity expires or must be recertified

The seven non-negotiable rules

The passport enforces seven rules that keep an agent from becoming an unmanaged privileged identity:

• No owner, no access
• No shared credentials — every agent has a unique identity
• Access must be temporary and least-privilege
• High-impact decisions require human approval
• Inputs (emails, files, tickets, web pages) are treated as potentially hostile
• Monitor behaviour, not only authentication
• Test the failure mode — the safest agent fails safely under hostile conditions

Agentic threat assessment built in

A dedicated section maps each agent against the OWASP Top 10 for Agentic Applications 2026 and common abuse cases, with the key control and test for each:

• Agent goal hijack and prompt / indirect-prompt injection
• Tool misuse and unexpected code execution
• Identity and privilege abuse (non-human identity risk)
• Memory and context poisoning
• Insecure inter-agent communication and unsafe delegation
• Cascading failures, runaway loops and rogue agents

Pre-production go / no-go gate

A one-page approval checklist gives security and risk owners a fast, evidence-based go/no-go decision before any agent reaches production — covering identity, least privilege, secrets, human approval for high-impact actions, monitoring, a tested kill switch and approved residual risk.

How it maps to your compliance obligations

Calling a system "AI" does not create a compliance exemption. The passport connects each agent to PCI DSS scope (where it can reach the cardholder-data environment), ISO/IEC 27001 and SOC 2 asset, access and change-management processes, DPDP Act obligations for any personal data it processes, and ISO/IEC 42001 AI-management-system governance — so one record evidences several obligations at once.

Frequently asked questions