IRVINE, Calif., Nov. 12, 2026 (GLOBE NEWSWIRE) — SecureAuth, a frontrunner in AI-driven identity security, today introduced Agent Checkpoints in Agent Authority, its control layer for AI agents. Teams can hand agents broad work without approving every step, and the actions that carry real risk still reach a person before they complete.

Human oversight for agents is often configured once, against a job role, and left in place. Overly restrictive settings lead people to stop using the agent, while loose settings leave it with broad permission that rarely gets reviewed. A fixed setting struggles because agent behavior varies. Given the same permissions and instructions, an agent can still act differently from one task to the next.

What a checkpoint asks for depends on what triggered it. In many cases the action goes ahead, and the person is notified afterward. In others, the system confirms that a real person is present and paying attention before it continues, without asking them to make a decision. Some actions need explicit approval, sent to the person’s phone, outside the agent’s own session. When the signal is that the agent is behaving unusually, the decision goes to whoever owns the system being touched, or to the security team.

Agent Checkpoints cover both human-in-the-loop and human-on-the-loop oversight and decide which one applies per call. A notification keeps a person on the loop without slowing the agent down. An approval puts them in it, before the action completes. Either way the loop closes at the moment of the action, rather than being configured in advance against a role.

“Agent Checkpoints decide on every request whether a person needs to be involved,” said Renjith Babu, Head of Product at SecureAuth. “Most of what an agent does goes through on its own. When a request looks unusual or meets set conditions, a person sees it in time to stop it.”

Checkpoints weigh several signals together, including who delegated authority to the agent and the agent itself. A checkpoint can fire on a spending or volume limit, unusual agent behavior, a system flagged as sensitive, or the details of the request. That last signal is where many controls stop short. Letting an agent use the payments system is a different decision from letting it send $50,000 to an account it has never paid, and only a control that sees the amount can tell them apart.

“Hardcoded limits based on tool name or user identity are cumbersome, but they were sufficient for human-driven sessions,” said Bryan Alsdorf, Vice President of IT and FinOps at MariaDB. “They do not scale to agent-driven workloads. SecureAuth evaluates multiple signals and brings a human in only where one is needed, which gives us the AI governance we need without laborious up-front setup or approval fatigue.”
Checkpoints apply whether an agent acts for an employee or a customer or runs on its own, across tool calls, model prompts, and AI use on the device. Decisions are logged against both the person and the agent, so there is a record of who authorized what.

“Enterprises want agents that can do meaningful work, and that means giving them meaningful authority,” said Geoffrey Mattson, Chief Executive Officer at SecureAuth. “The challenge is knowing when that authority needs a human check. Agent Checkpoints let businesses delegate with confidence, keeping agents moving while bringing the right person into consequential decisions.”

Agent Checkpoints are in early access for existing SecureAuth customers. New customers can request a demo and a 14-day trial of Agent Authority.

ABOUT SECUREAUTH

SecureAuth gives businesses continuous access control over what people and AI agents can do in the systems that run their business. The SecureAuth Continuous Authority Platform spans Workforce Authority, Customer Authority, B2B Authority, Presence Authority and Agent Authority, with a shared risk engine, so authority can be evaluated continuously rather than granted once at sign-in. SecureAuth is SOC 2 and ISO 27001 certified and OpenID Certified, and is headquartered in Irvine, California. Learn more at secureauth.com.

MEDIA CONTACT:
Kristen Grossi, talkTECH
Kristen@talktechcomm.com