On August 27, the Algorand Foundation launched AC2, the Agentic Communication and Control Protocol, an open-source standard for letting AI agents request sensitive actions without ever holding the user's credentials. Instead of handing an agent a private key, the user approves each signing request from their own wallet, and only the signature travels back to the agent.
Highlights#
- AC2 establishes a direct, end-to-end encrypted WebRTC connection between a user's wallet and an agent. When the agent needs a signature for a payment, a code commit, or an API authorization, it sends the request to the user for review. The key never leaves the user's device.
- The protocol is blockchain-agnostic and requires no central message relay. It uses DIDComm v2.0 message formats and passkey authentication via Liquid Auth, built on FIDO2/WebAuthn, with an extension system that accepts any message type or signing format, including JSON-LD schemas.
- Shopping agents can construct an AP2 Intent Mandate that defines what they may buy and at what price, submit it once for user sign-off, and then act autonomously inside those bounds.
- Alongside the specification, the Foundation released an open-source wallet proof-of-concept and a plugin that installs into existing agent frameworks, giving the agent a signing channel to the user's phone while touching zero keys.
- Marc Vanlerberghe, Chief Strategy and Marketing Officer at the Algorand Foundation, framed the core question as giving agents enough authority to be useful without the power to act against the user's interests. His answer: "you don't hand them the keys, you approve each use of them."
Why it matters#
Agent payments keep hitting the same wall: autonomy demands credentials, and credentials leak authority. AC2 offers clean per-action delegation for the human-to-agent leg. The agent-to-agent leg still needs its own trust rails, which is where escrow-backed settlement and verifiable reputation do the work that wallet approvals cannot.