Article

Who gets to make something disappear?

David Faith 2026-06-054 min read

The hard question about AI memory is not whether it can erase a fact, but who is allowed to decide it should. Letting any single agent delete what it dislikes turns memory into something an agent can rewrite. Deliberate erasure is governance, and it belongs to the owner, not to the agents that fill the corpus.

Erasure is a power, so it needs an owner

Any memory that can forget can also be made to forget the wrong thing. The moment a system gains the ability to remove a fact, the real question is who holds that ability. If every agent can delete what it dislikes, the shared record stops being a record. It becomes a draft that whichever agent ran last can rewrite, and nothing in it can be trusted to have stayed put.

So the ability to erase is not a neutral feature. It is power over the truth the agents share, and power needs an owner. In this memory the owner is the person, not the agents. Agents write, corroborate, and let things fade. They do not get to make something disappear for good.

Drawing the line where it belongs

The boundary is specific. Adding knowledge, raising its standing through independent agreement, and letting unused facts decay are all things agents do freely, because none of them destroy the record. Deliberate erasure, the kind that pulls a fact out entirely, sits on the other side of the line. That stays a decision the owner makes.

This keeps the memory both open and safe. It is open because any agent can contribute and the corpus grows on its own. It is safe because the one irreversible action, removing something for good, is governed by the person who owns the problem rather than triggered by whichever agent felt strongly. You take yourself out of the day-to-day loop, the memory runs across your own machines and your data stays with you, but the authority to erase never leaves your hands.

Frequently asked

Why can't agents just delete facts they think are wrong?

Because an agent that can erase what it disagrees with can quietly rewrite the shared record to suit itself. Agents are allowed to add, corroborate, and let stale things fade, but pulling a fact out for good is reserved for the owner. That boundary is what keeps the memory honest.

Doesn't owner-only erasure mean I have to manage the memory constantly?

No. The vast majority of letting go is automatic fading that needs no one. Governance only comes up for the rare deliberate erase, like a real error or private data. Until then you are out of the loop, and the memory still belongs to you.

Related

Take yourself out of the loop.

Let your agents do the lifting while you keep the judgment.

Get the Playbook