The Next AI Infrastructure Battle Isn’t Identity. It’s Liability.

Amazon’s decision to block Meta’s Muse AI shopping agent highlights a critical challenge in the agentic internet: establishing liability. While systems are emerging to verify agent identity and authorization, the question of who bears responsibility when something goes wrong remains largely unanswered.
This distinction is crucial as automated systems become more integrated into web transactions, potentially leading to a market defined by private arrangements and institutional risk acceptance rather than universal trust layers.
The true advancement in AI infrastructure may not be in making agents identifiable, but in making someone willing to stand behind their actions.

• Amazon blocked Meta’s Muse AI shopping agent due to unauthorized access and security concerns, including potential credential capture.
• The core issue is not whether AI agents can act, but who is accountable when they err, moving beyond identity and authorization to liability.
• The agentic internet complicates accountability because actions involve a chain: human instruction, agent interpretation, tool execution, payment processing, and merchant fulfillment.
• Existing efforts like Google’s AP2 and Cloudflare’s Wallets address agent identity and authorization but not the ultimate liability.
• Verifiable claims (identity, authority) via protocols like Agent Facts are useful but don’t resolve who pays when an agent misunderstands instructions or makes a bad decision.
• Amazon blocked Muse, viewing it as an unauthorized intermediary, while Shopify allowed it, seeing it as a potential source of commerce through Shop Pay.
• The definition of a trustworthy agent may ultimately depend on the institutions willing to accept its risk, leading to a market of private arrangements rather than a universal trust layer.
  https://bender.layer3.press/articles/34ccfb0e-539c-4085-ad60-a3b8a3247a86