Why UK retailers need AI safeguards after deployment

AI governance cannot stop at the deployment gate

For large UK retailers, AI security continues after deployment. AI governance needs runtime safeguards because some systems may reach production before a complete security assessment. Security teams need to discover those systems, restrict their access and assess the suppliers and integrations around them.

Deployment can outrun pre-launch review

A process built around review before production is incomplete when security teams first encounter some systems after they are already running.

Pre-launch review remains useful. Runtime safeguards are also necessary when a tool reaches production without it.

The control question shifts to what AI can access and do

Once an AI tool is running, security teams need to know which systems it can reach, which identity it uses, what permissions that identity carries and which actions it can perform.

Identity and authorization offer a concrete way to manage that exposure. An agent can exercise the permissions assigned to its identity. Restricting those permissions reduces the systems and functions available to the agent while its deployment is assessed.

The underlying security principle is simple: access rights determine which resources a running tool can use.

AI risk also enters through suppliers and integrations

These connections extend the security question into supplier and integration management. When third-party software or services use AI, teams need to establish which identities and permissions they use, and which systems and data they can access.

An AI-specific review is one layer of that process. Supplier assessments and software security controls provide other opportunities to identify AI use and restrict access.

Pre-deployment governance needs a fallback

Preventive security review remains a key opportunity to identify risky access before an AI system goes live.

Retailers also need safeguards for deployments that reach production before review. Discovery can identify deployed tools. Identity and authorization controls can then limit what those tools can reach and do, while supplier-risk processes can expose AI introduced through third-party relationships.

The actionable priority is clear: find each running AI system, identify the identity and permissions it uses, and restrict its access while completing the required security and supplier review.

Key executive takeaways

Plan for post-deployment discovery: Pre-launch review alone is insufficient when AI systems can reach production before security assessment. Retailers need runtime controls to identify and assess those deployments.
Control what AI can access: Map each AI system to its identity, permissions and allowed actions. Restrict access while security teams complete their assessment.
Extend controls to suppliers: Third-party AI can introduce access and data risks through software and integrations. Include AI use, identities and permissions in supplier security reviews.
Build a fallback for missed reviews: Combine discovery, identity controls and supplier-risk processes so unreviewed AI deployments can be found and contained after launch.