Access

Access defines who in the organization can build with AI, which identity they use, and how long they need to start. Delays here block the rest of the Build path.

Time to first change

Measure access by the time a new person needs to ship a small working change. A login is only one step in that path.

Without an enablement layer, tool licences, API keys, dataset access, and current documentation may arrive through separate queues. The elapsed time includes all of those waits.

The DORA program uses lead time for changes to measure delivery performance. The same measure can cover the path from an AI idea to a running change for each builder group.

Role-based grants

A manageable access model defines a small set of builder profiles, attaches tool and data grants to each profile, and assigns people through existing identity groups.

A useful starting set is:

• Engineer: full agent tooling, sandboxed execution, access to non-production data by default, production data by exception and with approval.
• Analyst or business builder: a supported low-code or notebook path, access to the data domains their role already covers, no ability to grant themselves more.
• Reviewer: read access to what others have built, plus the ability to approve the exceptions.

Profiles make access decisions reviewable. A new profile requires an explicit decision, while individual exceptions tend to accumulate without a coherent policy.

Fast, time-bound exceptions

When production data requires approval, the approved route must fit normal delivery timelines. A four-day approval process creates demand for a four-minute workaround.

Keep the exception path short, logged, and time-bound. Prefer a one-week grant with an expiry over permanent access reviewed annually.

Unsanctioned access

When the supported route is unavailable, people may use personal accounts and paste in the material they need. Treat this as an access-design problem and provide a usable sanctioned route.

Cisco's 2024 Data Privacy Benchmark Study, a survey of 2,600 privacy and security professionals, found that 48 percent of organizations admitted entering non-public company information into generative AI tools, while 27 percent had banned the tools at least temporarily. The figures show that bans and non-public data use coexisted.

Access decisions also affect shadow AI. Delays in the sanctioned route increase the number of unmanaged accounts and data flows that Secure teams must later identify.