Daon, a Fairfax, Virginia-based identity and biometrics company, announced this week that it has completed a three-part patented "trust stack" for controlling autonomous AI agents. The company's third and final agentic AI governance patent was awarded this week. The core problem the patent addresses is blunt: most enterprise identity infrastructure has no way to stop an AI agent mid-action. As the announcement frames it, if an agent makes a payment, changes a user's credentials, or executes a trade today, there's no mechanism to intervene once the session has started short of a human manually stepping in before it begins. Daon's answer is to move authorization from session-level to action-level. Rather than granting broad permissions at login and hoping the agent stays within them, the system checks each sensitive action individually, at the moment the agent attempts it. That's a meaningful shift from how most current agent deployments handle trust — a single OAuth-style login token that covers everything the agent does for the rest of the session. This lands in a crowded month for agent identity infrastructure. The Model Context Protocol shipped authorization hardening that enforces mandatory validation of the issuer parameter in its auth flow, closing a class of mix-up attacks where a client could be tricked into trusting the wrong identity server — described by its maintainers as preventive engineering rather than a response to an actual incident. Separately, the Decentralized Identity Foundation has been stewarding MCP-I (recently renamed KYA-OS), a framework using DIDs and verifiable credentials so a downstream service can answer who authorized an agent and what it's allowed to do, without prior coordination between parties. What ties these together is a shared diagnosis: authentication (proving an agent is who it says it is) and authorization (proving it's allowed to do the specific thing it's about to do) are different problems, and the industry has mostly solved the first while leaving the second to broad, static permission grants. Daon's patent, MCP's issuer-validation fix, and the DIF's credential framework are all attempts to close that gap from different angles — biometric/patent, protocol-spec, and decentralized-identity, respectively. None of this is theoretical hand-wraving. Real payment and credential-changing actions are the examples cited, and the timing — arriving as enterprises scale agent deployments faster than their identity systems can track them — is not coincidental. The harder question, still open across all three efforts, is whether per-action checks can run at machine speed without becoming a new bottleneck of their own.