Agent身份需四重ID：用户、模型、框架、运行时

An agent's identity is not one account. It takes four.

On @Vishwa_lab's Hubble podcast, our Co-Founder & CTO @scottshics sat down with Vishwa Co-Founder @nathan_sj_stem to break down what has to be pinned before an agent can transact.

▷ Four IDs, not one. User ID (who is behind this agent, and what their history looks like), model ID, agent harness ID, and runtime ID. An agent's behavior can only be judged once all four are pinned together.

▷ Why the model version matters. A year ago a model recommended sending a transaction to a spammer account on Solana, because its training data had been poisoned. Without a pinned model ID, you cannot tell which version did it.

▷ Why the harness matters. Claude Code and Codex behave very differently. Different philosophy, different logic. Combine that with the rest and they are different agents.

▷ The merchant side needs verifying too. Hand an agent your credit card or PII, and a malicious merchant can talk it out of them in seconds. Scott's framing: an agent behaves like a four-year-old. That makes it a KYB problem, not only a KYA one.

Pin the identity first. Behavior and payment both depend on it. 🪁