Observed arrival · 2026-11-05
Bollard: crash ratings for money-handling agents
Bollard presents a proposed rating system for AI agents that handle money, with ratings tied to a specific agent version and defined tests.
- For
- Builders, wallets, and researchers evaluating money-handling AI agents
- Worth noticing
- The page says each rating records one agent version's performance in defined tests, and explicitly says it is evidence, not a guarantee.
Field notes
The visible request form collects an agent name, version, ERC-8004 ID, and test endpoint, then offers to save a local draft; it says no request, payment, or wallet action is sent at that stage. Bollard describes a five-step sequence—challenge, simulate, attest, publish, check—and lists simulated wallets, attested execution, bonded quorum, and versioned ratings as components. Its displayed 300 USDC/300 USDG fee is labeled proposed, and the Robinhood Chain mirror is planned.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue