Agent Wormhole is a security layer for AI agents that transact with money. The project addresses a specific attack surface: AI agents read text (token metadata, tool descriptions, web pages, messages from other agents) and act on it, which allows hidden instructions in that text to manipulate an agent's behavior. Agent Wormhole provides several components: a launch layer that scans token launches on Robinhood Chain at creation and issues signed attestations bound to the token's exact metadata bytes; wormhole-guard, an open-source command-line tool on PyPI that protects AI agent configuration files from unauthorized modification; an x402 payment guard that verifies a proposed payment's recipient, amount, token and chain against the merchant's quote offline before the transaction is signed; and published security research with reproducible methodology. The WORM token is used for micropayments generated by the protection layer; fees fund buybacks and burns.