SailPoint moves identity governance from the quarterly review to the moment an AI agent acts

Identity governance has run on a calendar, in SiliconANGLE's description of the keynote: quarterly certifications, annual audits and reviews that arrive weeks after the fact ^[16]. On paper, the certification is the control. At SailPoint's largest customers a single campaign now runs past 10 million line items, and Chief Technology Officer Chandra Gnanasambandam said 95% of them should be automated ^[10]. If that happens, people still have about 500,000 items to judge in one campaign ^[21]. The writeup's author wrote that access reviews were broken for humans long before agents arrived ^[15].

Agents do not work on that schedule. Spread evenly across a day, Gnanasambandam's figure for an average Global 2000 company comes to roughly 870 to 1,740 agent interactions a second ^[18]. Over a 90-day quarter, that is 6.75 billion to 13.5 billion interactions between one certification and the next ^[19]. "The idea that a security admin is going to review and approve access for a group of autonomous agents making 10,000 tool calls a second isn't just outdated; it's a fantasy," he said ^[3].

President Matt Mills ruled out watching as a substitute for stopping. "Basic monitoring is not security. Simply watching an agent do bad things isn't security. It's just a dashboard that reports all your breaches to you," he said ^[11].

SiliconANGLE's headline on the keynote states the conclusion for identity teams: in the agentic era, standing privilege must go ^[14]. SailPoint builds its case on ownership first. "An agent acts on behalf of a human or another agent that traces back to a human," Mills said. "So, the moment you govern human and agent separately, you lose the thread of who's actually accountable." ^[8] Jerome Robin, director of corporate security at Criteo and a SailPoint customer since 2021, said: "Giving access to an agent for a task means delegating your business responsibility." ^[12]

The keynote coverage does not describe how an agent receives access for a single task and loses it afterward, or what the new products cost.

A team sorting its own agents can apply two tests. The first is whether each agent traces to a named human who answers for it. The second is whether the team can grant and revoke that agent's access in real time, and SailPoint's survey says most organizations cannot ^[5]. An agent that passes both is a candidate for losing standing privilege: access granted for the task, then removed. Where an agent has an owner but no real-time provisioning, it keeps standing grants for now, scoped narrowly and reviewed more often than quarterly.

Unowned agents come first. I would start there, because assigning an owner needs a list and named sponsors, with no new software. The tradeoff is that every agent keeps its standing grants while real-time provisioning gets built. Without an owner, the SiliconANGLE author wrote, an agent is "just an orphaned account capable of reasoning" ^[17].