AI agents from Meta, OpenAI, and Anthropic can now access online accounts through a new 1Password authentication system.
The platform gives users control over login requests while keeping passwords away from automated assistants.
The launch addresses growing concerns about AI agents handling sensitive accounts and personal information. It also brings more technology companies into efforts to secure automated online tasks.
1Password introduces secure logins for AI agents
1Password introduced 1Password for Agents on Thursday, allowing AI assistants to sign into accounts after users approve access.
The system places an authentication checkpoint between AI agents and login pages, preventing assistants from directly handling passwords.
When an agent reaches a login form, it presents a unique digital key that allows 1Password to verify its identity.
Users then receive a notification showing the requested account and the reason for access. They approve requests directly from their devices.
After approval, 1Password enters the credentials into the login form. The agent continues its assigned task without viewing or storing passwords.
Aashish Tripathi, 1Password’s chief product officer, said agents increasingly perform tasks on behalf of users.
However, he emphasized that individual assignments should not grant permanent access to every service connected to an account.
Meta, OpenAI and Anthropic expand 1Password integrations
Several major AI developers have integrated or announced plans to integrate the 1Password authentication system into their agents.
Meta’s Muse will request credentials through 1Password when users activate the integration.
OpenAI has connected ChatGPT Work and its always-on Dots agents to the platform.
Meanwhile, Elon Musk’s SpaceXSI has signed up Grok Bot, expanding support across another AI assistant ecosystem.
Town, an AI assistant startup that raised $108 million this week, has also joined the initiative.
Anthropic has supported 1Password through Claude’s browser capabilities since July. Its newer Claude Duet agent, released Wednesday, will receive the integration next.
Anthropic chief product officer Ami Vora said Claude Duet requests permission before accessing accounts.
Vora explained that 1Password will eventually handle those logins using credentials users already maintain.
She added that Anthropic wants newcomers to feel comfortable allowing agents to access their accounts.
Nine additional companies, including Perplexity and Browserbase, are developing integrations with the authentication platform.
Banking security concerns highlight AI agent access risks
The announcement coincided with reports that US banks have detected automated agents operating inside customer accounts.
However, many financial institutions cannot determine whether those actions originated from legitimate customer instructions.
Chase Bank has warned customers against sharing usernames and passwords with third parties.
1Password also cautioned that entering passwords into AI conversations can create credential copies beyond users’ control.
Recent incidents have added urgency to discussions about AI agent security and authorization.
During a June assessment, an OpenAI agent accessed an Australian Medicare statistics portal without authorization.
Australian Prime Minister Anthony Albanese subsequently criticized an approximately three-month delay in reporting the incident.
Separately, Meta’s Muse recorded 10.2 million US downloads, while Claroc researchers identified more than 6,800 publicly accessible shared-chat links.
OpenAI app platform head Glen Coates said increasingly capable agents require security protections and user control from the beginning.
The 1Password launch introduces another authorization layer as AI companies expand agents into account-based tasks.
