Most AI governance conversations focus on data and models. The next gap is already here and it's quieter: agent memory.
Agents remember. They carry context across turns, persist state across sessions and increasingly maintain long-term memory that spans users and tasks. That memory is what makes them useful and what makes them risky. Ask most teams what their agents retain, what they recall and whether one session's memory can surface in another's and you'll get a blank look. Memory is treated as an implementation detail, not a governed object.
It needs to be governed, for the same reason source data is. A memory entry can hold sensitive information. If it's recalled for the wrong user, that's leakage: cross-role or cross-session, with none of the access controls that protect the original source. Memory can also be poisoned: content written earlier can steer an agent's later behavior in ways no one reviewed. And because memory accumulates quietly, the exposure grows over time, invisibly.
Governing memory means treating each entry the way you treat a chunk of source data. It carries sensitivity and policy. Every write is checked: should this be retained, at what sensitivity? Every recall is checked: may this user, for this purpose, see this? And every operation is logged, so memory is auditable rather than a black box.
This matters most as enterprises move from single-shot assistants to long-running, multi-agent systems. The longer an agent operates and the more it remembers, the more its memory becomes a concentration of context and risk. A governance program that controls retrieval but ignores memory has a hole exactly where the most durable context lives.
The encouraging part is that memory governance isn't a separate system. It's the same discipline: meaning-aware policy, applied at runtime, with audit extended to one more surface. If you already govern retrieval and inference by role, intent and sensitivity, governing memory is the natural next step. The teams that take it now will avoid a class of incidents the rest are about to discover.