Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsPolicy

PCI SSC guidance: agents that can read cardholder data need a person to approve each action on it

The PCI Security Standards Council announced "Security Considerations for AI Systems" from Edinburgh on October 7, advisory guidance for payment environments written with its assessor roundtable and Board of Advisors; where it conflicts with a PCI standard, the standard wins. An AI agent with access to cleartext cardholder data should get explicit human approval for any action involving that data, and elsewhere the guidance allows either approval per task or "monitored autonomy" within set limits, with shutdown triggers, rollback and a named person accountable. It asks for each system's purpose, permissions and data access to be defined before deployment under a "least agency" principle and enforced by independent controls such as identity policies and network isolation, and warns against giving one AI system sensitive data access, outside communication and untrusted input together. It also covers an AI inventory and bill of materials, finding unapproved tools, adversarial testing before functional testing, keeping passwords and keys out of AI systems, treating tokenized data plus detokenization tools as readable data, and incident plans for prompt injection, model poisoning and actions outside approved scope.

Why it matters

The payment card standards body now says in writing what it expects from AI agents near card data. It is advice, not a requirement, but its line on splitting data access, outside communication and untrusted input across agents is a concrete test any payment team can run against its own agents today.

Sources

Read it with the rest of the news in the Learning Center