ai data.
Verified model outputs, signed at inference.
Models are the new data sources. Signal signs every inference — the prompt hash, the output hash, and the model that produced it — so agents can trust what another model said.
modelsopen + private
signedprompt · output · model
purposeagent trust
latency<200ms
trust between machines
An agent that reads another agent's output needs to know it wasn't tampered in transit. Signal gives every inference a receipt, so agents can reason about what they read.