Attestation
One Fortune 500 Company, 10,000 Unowned Agents
SailPoint's Navigate 2026 conference surfaced a number that puts the identity-sprawl problem in concrete terms: in a single Fortune 500 proof of concept, the company found more than 10,000 unknown agents running inside the enterprise with no clear owner attached to them. That's not a hypothetical risk model — it's a count from one customer's environment, discovered only after SailPoint went looking.
The broader data backs up why that number isn't an outlier. SailPoint's fifth annual Horizons of Identity Security report, released at the event, found that 79% of organizations run AI agents in production, yet only 2% use identity security tools to govern them, and just 15% can provision non-human access in real time. Most companies, in other words, have agents acting in production with essentially no governance layer watching what they do or who's accountable for them.
The framing that stuck out from the keynote: an agent without a human owner is just an orphaned account that happens to be capable of reasoning. A SailPoint customer put the stakes plainly — giving an agent access to perform a task means delegating business responsibility, whether or not anyone formally signed off on that delegation.
SailPoint's answer is a shift from kill switches to attribution. Security experts speaking at the event argued that agents need attribution and identity binding, not human-style anonymity — every action has to trace back to a responsible human or organization from the outset, not just get shut off after something goes wrong. The company's new tooling (Agentic Fabric, Autonomous Agents, and an Agent Audit feature) is built around that idea: continuous discovery of shadow agents, runtime authorization that checks what an agent is doing at the moment it acts, and exportable audit evidence for when regulators or auditors ask who owns what.
What this says about the state of the industry is blunter than most vendor announcements admit: the problem right now isn't exotic attacks on agent protocols. It's that most large enterprises don't actually know how many agents are running inside their own systems, let alone who's responsible for them. Discovery — the boring, unglamorous work of finding what already exists — is turning out to be the precondition for every other piece of agent governance infrastructure, from runtime control to audit trails. You can't attest to what you can't first count.