1Password Opens Agent Sign-Ins Without Exposing Passwords

Under the system, an agent presents a unique digital key to 1Password when it reaches a sign-in form. 1Password then notifies the user about which account is requested and why. The user approves or rejects the request on their own device, a process The Verge compared with two-factor authentication. If approved, 1Password fills the login information into the page rather than passing it to the agent.

Engadget reported that access is granted per session, limited to specific approved items and not carried over. Passwords and one-time MFA codes are injected through a secure channel managed by 1Password, keeping them outside the model. 1Password also allows authenticated multi-site sessions to complete multi-step workflows without prompting the user for credentials at every step.

“Agents are starting to do real work on our behalf. The old model where you hand over a credential and hope for the best doesn’t work,” Aashish Tripathi said in a statement cited by The Verge. The Verge identified Tripathi as 1Password’s CEO; SiliconANGLE and Engadget described him as chief product officer.

Launch partners include Meta’s Muse, OpenAI’s ChatGPT Work and Dots, Claude Duet, Grok Bot and Town, according to the reports. 1Password said Browserbase, Browser Use, Dana, Instinct, Kernel, Perplexity, Ramp, Tomo and Wajo are working on additional integrations. SiliconANGLE reported that Claude has worked with 1Password in the browser since July and that Perplexity and Browserbase browser agents have been connected since late 2025. 1Password did not say when each partner integration would reach users; Engadget reported that pricing and availability details had not been disclosed.

The launch follows a push by 1Password for task-based agent permissions. At Okta’s Oktane conference last month, 1Password Chief Technology Officer Nancy Wang said an agent should have to prove it completed its last task at the right permission level before receiving more access, according to SiliconANGLE.

On the same day, startup Action State Group Inc. released open-source tools for verifying records of what AI agents did as they buy, sell and delegate work across company boundaries. The offering includes EvidenceBook for tamper-evident, signed transaction histories; Terms Packs for evaluating agent behavior against rules; and Countersign, an independent verification service. Action State said participants can compare selected records without a central authority holding everyone’s data. It said EvidenceBook can verify that reported events occurred but cannot establish that an agreement was satisfied, and Countersign does not certify claimed outcomes or completeness. Terms Packs do not certify legal compliance, the company said. EvidenceBook is available under Apache 2.0; Action State’s specifications remain Internet-Drafts rather than IETF standards. Countersign is currently free, and the company plans to charge once a relying party accepts its signatures.