1Password is an AI agent that manages account access with user approval. 1Password for Agents announced its platform. According to the company’s statement dated October 22, supported agents can request permission to use the credentials in the vault when logging into a service is required. If the user approves, 1Password fills in the login information on the target page itself. This way, there is no need to send the password to the agent in chat.
The purpose of the new system is to separate the intermediary from being able to do a job and having access to the entire password vault. The request indicates to the user which account needs to be logged in. The user selects and confirms the relevant record, and the agent continues its task in the opened session. The company states that this flow proceeds without the password and single-use code being visible to the artificial intelligence model.
Despite the initial announcement, not all partners are ready for launch on the same day. There is a distribution statement for Meta Muse and Grok Bot integrations in the coming days. Support for Claude Duet, ChatGPT Work, dots and Town Townie is listed as coming soon. Therefore, just because the partner name is in the announcement does not mean that it will work on your account today.
Availability also depends on account region. The platform is initially offered to Individual and Family accounts on 1Password.com at no additional charge. 1Password.eu and 1Password.ca accounts are currently excluded. Users can disconnect, but supported platforms and access controls may vary depending on the particular broker integration.
1Password requires user confirmation and account region
On the other hand, the company associates this approach with its confidential computing security model. The model executes certain operations in the cloud in protected environments separated at the hardware level. The client verifies the identity of this environment and the code it runs with cryptographic evidence. The company states that server administrators cannot directly access the memory of these environments.
According to the security document, each code version running in the protected environment leaves a trace in the transparency log operated by the third party. The client checks the code signature, the published record, and the metrics of the environment together. When one of these checks fails, it rejects the connection. These are architectural descriptions published by the company, not a standalone security guarantee that covers every use case for the new brokers platform.
The fact that the password is not visible to the model does not eliminate the question of what the agent can do within the account you approve. 1Password’s announcement emphasizes only approving logins required for the desired job. The company also makes clear that protections vary by integration. For this reason, it is necessary to examine the scope of both the selected account and the task requested by the broker.