Cool use case — the trust layer for AI-generated content is becoming critical. How are you handling false positives?
By false positives, I'm assuming you mean, two separate LLMs/agents output and sign the exact same content (e.g. hello world!).
These are not traditional signatures and can have metadata encoded into them (the demo only encodes lease ID, create timestamp). So you could rely on the metadata during verification for provenance (which one came first). The demo has a separate link to additional information about the signatures.
P.S. this is just a demo to demonstrate this in a real-time streaming scenario (mostly for a prospect). But we'd love to hear if anyone thinks this could be useful as a product of its own.