1Password adds user-approved logins for AI agents
Illustration, not documentary evidence of the event.
1Password has launched 1Password for Agents, a tool that lets users approve an AI assistant’s login without giving the assistant their credentials, according to techcrunch.com. The reported workflow puts the password manager between the agent and the account: the agent identifies itself using a unique digital key, the user receives a login alert, and 1Password fills in the credentials after approval.
The company frames this as a way to delegate tasks without leaving copies of passwords or other secrets with an assistant. That addresses a specific decision for users: whether an agent needs to receive a password at all. Under the described system, users can authorize a sign-in while keeping the credentials out of the agent’s possession.
The more consequential distinction is between approving entry and approving actions after entry. The reported identity check and user confirmation establish a path for authorizing a login. They do not, by themselves, establish which subsequent actions the user intended. For someone choosing how to delegate an account task, those are separate acceptance criteria: keeping the password private and constraining what the assistant may do once signed in.
That distinction also connects the launch to the banking concerns in the report. Techcrunch.com cites Axios reporting that banks are detecting agents inside customer accounts but often cannot determine whether their activity reflects customer requests. Read together, the two reports suggest that user-approved authentication and a bank’s ability to recognize authorized activity are separate problems. The described 1Password workflow should not be treated as evidence that the receiving service can recognize that approval.
Before approving a login, define the intended task and its stopping point, such as retrieving information versus submitting a change. Then establish whether the chosen agent and account service can enforce that boundary, including any required confirmation before a consequential action. If that cannot be established, keep that action under direct user control. Password confidentiality is a useful criterion for selecting an access method; it is not enough evidence to approve every task that access makes possible.