1Password's own research found that 40% of developers give AI agents persistent access to systems or credentials, the kind of standing access security teams have spent years trying to get rid of for human accounts. On July 28, the company launched Privileged Access to close that gap, extending its Unified Access platform into privileged access management for both people and agents.
What Privileged Access actually does
The core idea is just-in-time, least-privilege access. Instead of a service account or API key that sits there indefinitely, an identity (human or AI agent) requests access, gets a role scoped to the specific task, and loses that access automatically when the task ends.
For AI agents specifically, this is packaged as AI Agent Privilege Control. Every agent gets its own distinct identity rather than sharing a credential with other agents or services. The access is provisioned at request time and deprovisioned when the task finishes, and the agent itself never sees the underlying credential; it works through the broker instead.
Alongside this, 1Password put 1Password Credential Broker for GitHub Actions into public preview, aimed at the same problem in CI/CD pipelines where long-lived secrets tend to pile up.
How this connects to 1Password for Claude
This isn't 1Password's first move here. On July 16, the company announced 1Password for Claude, which added Agentic Mode and a zero-exposure framework so Claude can use approved credentials without the model ever seeing the secret itself. Privileged Access is the broader infrastructure play behind that: the credential-broker pattern extended to any agent, not just Claude, and to infrastructure access generally, not just password retrieval.
Pricing and availability
Privileged Access is priced per seat, and 1Password says it will sell the products inside Unified Access separately rather than forcing customers to buy the whole bundle. Credential Broker is in public preview now, available to anyone on a business-level Enterprise Password Manager plan or higher. Exact per-seat numbers for Privileged Access itself weren't published at launch.
| Standing access (old way) | Privileged Access (new) | |
|---|---|---|
| Credential lifespan | Persistent, often indefinite | Provisioned per task, expires automatically |
| AI agent identity | Often shared with other services | Distinct identity per agent |
| Credential exposure | Agent or script often holds the secret | Broker handles it; agent never sees it |
| CI/CD secrets | Long-lived tokens in pipelines | Credential Broker for GitHub Actions (public preview) |
Who this is actually for
This is not a tool an individual developer signs up for on a Tuesday afternoon. It's aimed at security and platform teams at companies already running 1Password's Enterprise Password Manager, who are trying to get ahead of the mess that comes from every team spinning up its own coding agents and integrations, similar in spirit to what [BridgeApp is trying to solve](https://questloops.com/blog/bridgeapp-s-new-orchestration-layer-wants-to-end-tool-switching-for-ai-dev-teams) on the workflow side rather than the credentials side.
The honest limitation here: adoption depends entirely on whether teams bother to wire their agents through the broker instead of just handing them an API key, which is faster in the short term and exactly the habit 1Password's own research says 40% of developers have already fallen into. A good product doesn't fix a bad habit by itself.
FAQ
**Does Privileged Access replace 1Password's regular password manager?** No, it extends the Enterprise Password Manager and Unified Access platform with PAM-style temporary access; it's an add-on for teams that already use 1Password.
**Is this only for AI agents?** No, it covers human privileged access too, but AI Agent Privilege Control is the part built specifically for agent identities.
**What does it cost?** Priced per seat, sold separately from other Unified Access products. 1Password hadn't published exact numbers as of launch.
