Three separate announcements this week show the AI agent identity stack solidifying into distinct layers: who the agent is, what it's authorized to do, and how it pays. The most concrete move: the Linux Foundation announced the operational launch of the x402 Foundation, an open-governance body created by Coinbase to steward the x402 protocol. As the release describes it, x402 embeds secure payment capabilities directly into web interactions so that AI agents, APIs, and applications can send and receive payments as seamlessly as they exchange data, with support for payment types ranging from traditional cards to stablecoins. The foundation already counts 40 member organizations including AWS, American Express, Circle, Google, Mastercard, Stripe, and Visa, according to the announcement covered by Biometric Update. That payment layer only matters if the agent making the purchase can first prove it's acting with real authority. Two other developments this week address that upstream problem. 1Password published details on its integration with Anthropic's Claude, built on what it calls a zero-exposure architecture: Claude can complete browser tasks requiring logins and one-time passcodes, but credentials never enter the model's context. 1Password's CTO Nancy Wang framed the design principle directly — the goal is to let a user give an agent permission to use a credential without letting the agent see it, so the agent knows it used a login without ever holding the password. Separately, Japan's Dai Nippon Printing (DNP) said it has begun offering a digital identity function for AI agents making proxy purchases in online shopping, built on its CATRINA distributed ID platform and using verifiable credentials. The stated aim is to create an environment where only AI agents authorized by the user can conduct transactions based on the user's will and authority. None of these three efforts overlap much technically — one is a payments protocol, one is a credential-injection product, one is a VC-based identity platform — but they're solving adjacent parts of the same problem: an agent transacting on a human's behalf needs to be identifiable as delegated, scoped to a specific task, and able to pay, without collapsing into an unrestricted proxy for the human's full identity. That separation of person, agent, delegated authority, and transaction is what's new here; each piece has existed in isolation before, but treating them as distinct, composable layers with named standards bodies behind each one is a shift toward production infrastructure rather than one-off integrations. The open question these announcements don't resolve is auditability across the stack — if a payment clears via x402 using authority granted through DNP's system and credentials injected by 1Password, no single party in that chain currently produces a unified, verifiable record of the full delegation path. That's the piece still missing.