Close the agent governance gap: Control access, track outcomes

Brought to you by Limina

Broadcast date: 13 Nov. 2026

Time: 08:00–09:00 PST, 11:00–12:00 EST, 16:00–17:00 CET

Somewhere in your organization, an agent opened a customer record to answer one question. It may have pulled more data than the task needed, passed some to a model, folded it into a summary, and spawned new records, embeddings or memories from the result. Most governance programs can tell you who was granted access to the source system. Far fewer are prepared to say what the agent retrieved, which controls applied, where the data went, and what was derived from it. 

That's the agent governance gap. An agent's actions are decided at runtime: what it retrieves, which tools it calls, what it combines, what it creates. Design-time review remains necessary, but it can't govern each execution on its own. 

Closing the gap takes controls at two points. Before the agent receives data, sensitive information can be discovered and transformed so the agent receives only what the task requires, in the least identifying form the approved use allows. As data move through the workflow, their purpose, permissions, provenance and applicable restrictions must travel with them, including into the summaries, embeddings, memories and other derivatives the agent creates.

Key takeaways:

• Why access approval alone does not determine how an agent will use data at runtime.
• How discovering and transforming sensitive data before agent access can reduce exposure while preserving useful context.
• What provenance and lineage records need to capture as agents retrieve, combine and derive information.
• Which controls organizations can implement today and which governance capabilities are still emerging.