Verify
Checks which agent is calling, who built it, and whose behalf it is acting on, before your systems act on its request.
The problem
An agent's description of itself is just text. A2A lets providers sign their Agent Cards (a JWS signature over a canonicalized card), which proves the card came from whoever holds the key.
A business receiving a request still needs to know whether that key belongs to a reputable provider, and whether this request comes from a real customer's agent or from an automated abuse campaign.
How it works
- Validate the signature on the agent's request and the key behind it.
- Match the key to a known provider record.
- Assess the request: history, rate, and reported abuse.
- Return a verdict your policy uses: allow, ask for more proof, or refuse.
Open standards
Verify checks HTTP Message Signatures (RFC 9421) as the IETF Web Bot Auth draft profiles them, alongside A2A's signed Agent Cards, and adds provider identity and reputation on top of the signature.
Availability
In development. A first version runs in Front Door: known agent providers, each business's own block and allow lists and history, and allow, ask-for-more-proof or refuse verdicts.
Questions
- Does Verify authenticate our customer?
- No. Verify is about the agent. Proof that the customer authorized an action comes from Mandate.
Next module: Mandate