1Password lets personal AI agents sign in with a person's approval and reported $500 million in annual recurring revenue. For employers, the open question is what happens to access an employee approved.

Two announcements from 1Password arrived on October 22. 1Password for Agents lets personal artificial intelligence, or AI, agents sign in with a person's approval (1Password, “Agents”), and the company reported more than $500 million in annual recurring revenue, or ARR (1Password, “Surpasses”).

The person approves each sign-in on their own device

The agent uses a unique digital key to identify itself to 1Password. The person sees which credential the agent needs and why, and approves or declines. 1Password fills the credential into the form, and the agent continues the task without receiving the password (1Password, “Agents”).

An agent an employee approves can sign in to a work account

A password kept from the agent protects the credential. An employer's ability to audit and end the agent's access is a separate requirement.

• Individual and Family accounts hosted on 1Password.com get 1Password for Agents at no additional cost. Accounts on 1Password.eu or 1Password.ca are not supported in the launch terms (Torrey).
• Business customers supply more than 75% of ARR, and 1Password points larger organizations to Unified Access for dedicated workplace AI assistants (1Password, “Surpasses”).
• 1Password says the need for account access covers work tasks as well as personal ones, and Town's chief executive says a company needs to know that access is safe (1Password, “Agents”).

An individual can disconnect an integration at any time (Torrey). I found no statement in 1Password's launch materials on:

• whether disconnecting ends a session the agent opened before the disconnect
• whether a security team can see or end an approval an employee gave through a personal plan

The August post on Privileged Access sized persistent access with 1Password's survey, in which 40% of developers said they give agents persistent access (Atwell). My May analysis of Cisco's agent security strategy made the lifecycle point that agents do not retire, so access piles up unless an event ends it. An employee leaving is such an event.

Five featured partners, with rollout stages that vary

1Password features five agent partners (1Password, “Agents”). Rollout stages vary, and safeguards and controls vary by integration (Torrey):

• Muse from Meta: rolling out over the coming days
• Claude Duet from Anthropic: coming soon
• ChatGPT Work and dots from OpenAI: coming soon
• Grok Bot from SpaceXSI, the former xAI: rolling out over the coming days
• Townie from Town: web and iOS versions coming soon

1Password lists nine more partners as coming soon, including Browserbase, Perplexity, and Ramp (Torrey).

The five-times growth claim comes without a defined comparison

1Password reports these figures (1Password, “Surpasses”):

• Net new ARR grew about 60% through the first three quarters of fiscal 2026.
• Free cash flow margin is double digit.
• Customers spending more than $100,000 a year grew more than 40% year over year.
• Gross retention held above 90%.

1Password associates its developer and AI security tools, used in 16% of business accounts, with about five times faster ARR growth among customers who use them (1Password, “Surpasses”). ARR is 1Password's own recurring revenue, so the claim concerns revenue from those customers. 1Password does not define the comparison group or show that adoption caused the difference, and fast-growing customers can adopt new tooling for reasons unrelated to the tooling.

Enterprises need to verify two things before a pilot expands

Audit trails have to bridge personal approvals and the company's records. Snowflake's Chief Security and Trust Officer, Mayank Upadhyay, says in the growth release that the Snowflake integration ties each agent action to the agent and to the person it acts for, which gives security teams a full audit trail (1Password, “Surpasses”). A log of login approvals and action-level attribution record different things, so 1Password would need to show which one covers personal agent approvals.

I found no availability stage for the Snowflake integration in 1Password's October materials, and my August post recorded it as scheduled for the second half of 2026.

Approval volume needs a policy. On October 20 I covered Amazon Web Services' argument that people reviewing a long approval queue stop catching the mistakes that matter. That argument concerns rules enforced over an agent's actions, and a policy that pre-approves routine logins is a narrower version of the same idea. Its suitability depends on its scope and its exceptions.

Keeping the password from the agent protects the credential. The access the agent holds is a separate question for the employer.