When an AI agent submits a checkout request on behalf of a user, the business receiving that request faces an information asymmetry problem. The platform — whether it's Microsoft Copilot, Google's Shopping Agent, or another commerce AI — observes rich transaction context: IP address, user agent, device fingerprints, behavioral patterns, authentication signals. The business sees none of this directly.
This matters for fraud prevention, rate limiting, and abuse detection. Without access to environmental signals, businesses can't make informed authorization decisions. They're forced to either:
None of these options work at scale for agentic commerce.
PR #203 introduces a signals contract that enables businesses to request environmental data from platforms, and for platforms to provide cryptographically attested observations. The key design principles:
Signals are platform attestations. Values MUST reflect direct observations by the platform, not relayed buyer claims. This is what makes them trustworthy for authorization decisions.
The flow operates in two directions:
risk and abuse codes to indicate what data they needThis creates a dynamic, capability-based system. Businesses declare what they need; platforms provide what they can observe. Extensions can define proprietary signals for specialized use cases.
The specification defines signal categories that platforms may support:
The feat! prefix indicates breaking changes. Specifically:
risk and abuse messagesFor platform developers, this requires implementing signal observation and attestation capabilities. For businesses, it requires parsing signal requests and incorporating provided signals into authorization logic.
The PR adds 181 lines and removes 63, indicating substantial new functionality with some consolidation of existing code. Key additions:
// Business requests signals via info message
{
"code": "risk",
"signals": ["ip_address", "device_fingerprint", "auth_method"]
}
// Platform provides signals on next request
{
"signals": {
"ip_address": {"value": "...", "confidence": 0.95},
"device_fingerprint": {"value": "...", "confidence": 0.87}
}
}
With signals in place, businesses can implement sophisticated authorization strategies:
Critically, this works without requiring user-facing authentication. The platform observes; the business decides.
This PR complements other recent UCP security work:
Together, these form the trust infrastructure that enables businesses to accept AI-driven transactions with confidence. Without them, agentic commerce remains a demo. With them, it becomes enterprise-ready.
Ilya Grigorik continues to drive the critical security infrastructure that UCP needs for production adoption. Authorization and abuse signals solve a real problem — the information asymmetry between platforms and businesses — in a way that respects privacy while enabling trust.
The breaking change is worth it. You can't have secure agentic commerce without platform attestations.
Follow the implementation: PR #203 on GitHub