Okta Wants To Secure Your AI Agents, Too

At its annual Oktane user conference, identity management service Okta today announced its plans to start treating AI agents — or nonhuman identities — as first-class entities on its platform. The company argues that its experience in identity and access management translates well to AI agents and that, as CEO Todd McKinnon said during his keynote, “to get AI agent security right, you have to get identity right.”

McKinnon argued that it takes an end-to-end approach to fully secure agents and ensure they only get access to the data they are meant to see. “It’s got to be comprehensive. It’s got to cover everything — no gaps, no little wedges for any kind of threat to speed through. It’s got to cover every data set, every use case and every resource,” he said.

Okta is branding its approach as an “identity security fabric,” with the actual product being Okta for AI Agents.

To put these ideas into practice, Okta announced a number of products and initiatives. Part of this involves getting AI agents into the Okta platform and making it easy to build agents with these security primitives built in. Okta is also pushing the industry to adopt the Cross App Access (XAA) protocol to standardize how AI agents and applications connect securely and is building it into its own agent-related products, as well as those from its Auth0 developer platform.

“There are a lot of standards in the AI world, but there’s a missing standard here, and that’s why Okta is working with the standards bodies to propose a new standard we call Cross App Access,” McKinnon said. “It’s focused on security and access, it lets IT and security teams set the access policies up front for these AI agents, which makes it open and transparent and visible to everyone involved.”

Okta for AI Agents will roll out in a number of different phases. It will include tools for discovering AI agents that are already in use in an organization and identifying the risk they pose. (Often, an agent may not be in use anymore but still connected to a long-living OAuth token or API key that a hacker could abuse, for example.)

Okta will then also start including agents in its Universal Directory, its database where, until now, admins could manage all of the people, groups and devices under their control. This directory will now also include AI agents, including information about what they should have access to, who owns them and more. And when things go wrong, IT and security teams will now have a way to log agents out of all of their access.

McKinnon stressed that Okta is doing all of this without telling businesses which platforms they should use to build their agents.

“[There are] over 8,000 integrations in our Okta integration network. We don’t have an opinion on which technology you choose. We focus on identity and leave the choice of technology up to you. There’s no lock-in,” he said. “We focus on the fundamentals — the fundamentals of identity, governance, visibility, control — and free you to choose whatever emergent technology in this dynamic landscape that serves your needs the best.”

At the core of all of this is the idea that businesses should essentially treat agents in the same way as they treat human identities on their platforms. Given that Okta already does that for its customers, it makes sense for the company to position itself to now also cover nonhuman entities.

We’ve seen similar arguments from the likes of Workday, but in the context of extending that platform from managing people and money to also include agents.

For the most part, I think the analogy works, but if the AI optimists turn out to be right, the next question will be about how these tools can scale as businesses bring on hundreds or thousands of agents.