1Password launches a trust layer for AI agents, and its own growth numbers back the bet

1Password launched a product that lets personal AI agents sign into real accounts without the agent ever seeing a password, and it paired the release with growth numbers that turn machine identity from a security talking point into a measurable revenue line.

A trust layer, not a password manager for bots

1Password for Agents went live on October 22. It works with named integrations across the current field of personal assistants: OpenAI's ChatGPT Work and Dots, Anthropic's Claude Duet, Meta's Muse, xAI's Grok Bot, and Town's Townie, with Browserbase, Perplexity, Ramp and several smaller agent builders also connected. When one of these agents reaches a sign-in form, it sends a unique digital key to 1Password to verify its own identity. The person gets an alert on their own device naming the exact credential requested and why. Only after they approve does 1Password fill the login directly into the form. The agent finishes the task; it never holds the password.

"1Password provides the trust layer that makes this possible across personal agent experiences."

That line, from chief product officer Aashish Tripathi, is the company's own framing, and it is worth taking at face value: 1Password is positioning itself as infrastructure that other companies' agents have to pass through, not a feature inside its own app. Chief business officer John Torrey made the ambition explicit too, saying personal AI agents have the potential to help millions of people get more done. The company says it already protects more than 1.6 billion credentials and secrets across more than 200,000 businesses, which is the install base it is now routing agent traffic through rather than building a separate product from scratch.

The number that matters more than the launch

The more useful piece of news arrived in a separate release the same day: 1Password said it has passed $500 million in annual recurring revenue, with net new ARR up roughly 60% through the first three quarters of its fiscal 2026 and Q3 ARR growth near 90% year over year. Net new logos rose 150% year over year in the same quarter. None of that is independently audited, so treat it as the company's own disclosure. But the detail that should interest software marketers is narrower: customers who use 1Password's developer and AI-security tools, the products built to manage machine identity rather than human logins, grow their ARR about five times faster than the rest of the customer base. Monthly active use of those tools grew 200% year over year, and the count of non-human identities active inside them grew 500% in the same period.

CEO David Faugno tied the two releases together directly: "AI agents are rapidly changing how people work and live, but unlocking their potential requires trust." Snowflake's chief security and trust officer, Mayank Upadhyay, is quoted making the same point from the buyer's side, arguing that engineering teams should not have to choose between shipping agents fast and shipping them safely. An IDC analyst quoted in the release put it more bluntly: identity is becoming the control point for agentic AI, not the agent and not the application it calls.

Who this actually affects

This is not only a story for security vendors. Any software company whose product will eventually be operated on a customer's behalf by an agent, rather than clicked through by a person, now has a credential-handling question it didn't have a year ago. Right now, most vendors answer it with a vague claim about being "agent-ready." 1Password answered it by naming five specific assistants by brand and publishing exactly how the handoff works, which is a much harder claim to fake and a much easier one for a buyer to check.

It also affects how security and IT buyers read a vendor's roadmap. Upadhyay's comment, that engineering teams should not have to trade shipping speed for safety, describes the exact tension a buyer is now screening for in any vendor that touts agent support: did you build a real approval layer, or did you just add an API key field and call it agentic. A product page that cannot answer that question in plain language is going to lose the comparison to one that can, regardless of which product is actually more capable.

What to do differently

• Write a public page describing exactly how an AI agent can and cannot access your product, and link to it from wherever security questionnaires land.
• Name the specific agents and assistants you've tested against, rather than claiming generic "agent compatibility."
• If you sell identity, access, or security tooling, start reporting non-human identity usage as its own line, since the market is already reading it as a leading indicator, not a footnote.
• Assume procurement conversations will soon ask whether an agent can use your product without ever holding a password, even for vendors with no agent features of their own.