GitHub updated Copilot Memory on May 26, 2026, pushing AI memory from a convenience feature toward a capability that needs explicit management. Copilot Memory remains in public preview and is available across paid Copilot plans, but the new controls address issues companies will encounter in real deployments.
The first change is clearer deletion guidance. When a user asks Copilot to forget something, Copilot now points to the right place to remove the memory and down-votes the memory where voting is available. This sounds small, but it matters for trust. If a system can remember preferences or repository facts, users need to know how to inspect and remove those entries.
The second change is a repository-level off switch. Repository admins can now disable Copilot Memory for a repository through existing Copilot feature controls in repository settings. Once disabled, repository-level facts are no longer stored or read. GitHub also notes that existing facts are not automatically deleted, and user-level preferences are unaffected.
The third change is a /memory command in the Copilot CLI. Users can run /memory on to enable memory, /memory off to disable it, and /memory show to check the current state. The choice persists across sessions, which is practical because AI agents increasingly move between terminal and IDE workflows.
The fourth important change is clearer scope at capture time. The store_memory permission prompt now states whether an entry will be a user-level preference or a repository-level fact. The former is visible only to the individual user across repositories, while the latter can be used by contributors on the repository. That boundary matters because personal preferences and shared team facts need different governance.
For AI agents to become useful in team work, memory is almost unavoidable. Without memory, the agent has to relearn repository conventions, architecture, style, and process every time. With memory, companies must manage retention, permissions, incorrect memories, shared visibility, and deletion responsibility. GitHub's update is part of that missing governance layer.
The signal for companies adopting AI coding tools is direct: the future governance surface is not only prompts, models, and tool permissions. It also includes what the AI remembers, who approved it, where it is visible, when it can be disabled, and how it can be removed. AI memory is becoming part of compliance and operations, not only user experience.
