Anthropic partnered with a corporate identity provider for an early-stage programme, letting joint customers govern assistant use and access to tool providers through a single identity layer.
The problem it addresses is familiar to anyone administering a corporate environment. An AI assistant connected to internal tools creates an access path bypassing the controls built over years for people and systems.
Tying an agent to the same identity provider used for employees resolves part of that, because it brings the agent's access into the existing inventory: who authorised it, what it can reach, when it expires.
A security company's per-action authorisation announcement in the same period pointed the same way, and both answer what the month's incidents showed: an agent inheriting broad permission reaches data nobody explicitly approved.
For operations without that kind of contract, the principle still applies simply: the agent should have its own identity, with its own permissions, rather than using a person's credentials.