1Password on Thursday introduced a platform that allows personal AI agents from Meta, OpenAI and Anthropic to sign into a user’s accounts after the user approves the request.
The agent executes the task without seeing or storing the password at any time.
Nine more companies, Perplexity among them, are building integrations
The product, called 1Password for Agents, adds a check between the agent and the login page. The agent taps into a sign-in form, presents a unique digital key, and 1Password authenticates its identity.
The user will be prompted with the account requested and why. Approval is on the individual’s own device.
Then 1Password fills in the credential directly into the form. The agent proceeds with the task, and the password stays in the vault.
“Agents are starting to do real work on our behalf,” said Aashish Tripathi, chief product officer at 1Password. He added that one task assignment shouldn’t give an agent permanent access to everything behind an account.
When the integration is on, Meta’s Muse will seek logins through 1Password. OpenAI wired up both ChatGPT Work and its always-on Dots agents.
Elon Musk’s SpaceXSI signed up Grok Bot. An AI assistant startup named Town, which raised $108 million this week, signed as well.
Since July, Claude has been using 1Password in the browser. The more recent Claude Duet agent from Anthropic, which was released on Wednesday, comes next.
“Claude Duet asks before it signs in anywhere, and soon 1Password will complete the sign-in with the passwords people have already set up, so they don’t have to start over,” said Ami Vora, chief product officer at Anthropic.
Vora said that Anthropic wants everyone to be able to use the feature, even those “who’ve never before let an agent near their accounts.”
Nine more companies, including Perplexity and Browserbase, are building their own integrations.
US banks now detect automated activity inside customer accounts
The launch came on the same day that reports emerged of US banks observing activity by agents inside customer accounts. Most banks can’t tell whether that activity links back to a real user request.
Chase Bank told customers not to give their usernames and passwords to anyone. 1Password said that typing a password into an AI chat saves a copy the user can’t control. Its new system is built to stop that.
During a June assessment, an OpenAI agent accessed an Australian Medicare statistics portal without authorization, Cryptopolitan reported. Prime Minister Anthony Albanese has directly complained about a delay of about three months in disclosure.
Meta’s Muse logged 10.2 million US downloads, but Claroc researchers also identified more than 6,800 shared-chat links anyone could access.
“As agents take on more complex tasks for users, security and user control need to be built in from the start,” said Glen Coates, head of app platform at OpenAI.
The smartest crypto minds already read our newsletter. Want in? Join them.