Show HN: AI Content Attestation & Verification w/ Revocable Signatures(lyfe.ninja)
lyfe.ninja
Show HN: AI Content Attestation & Verification w/ Revocable Signatures
https://lyfe.ninja/news/#know-your-agent-with-blkbolt
Link to demo in article. The article is for context.
4 コメント
Cool use case — the trust layer for AI-generated content is becoming critical. How are you handling false positives?
Thanks, we think so too, we're not aware of anything doing this currently but please share if you know of anything.
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.
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.
So it seems like this is mostly to tie content back the original agent, but doesn't actually do anything to verify the agent. Is that right?
True, this demo does not verify the agent or it's owner, but that's not necessarily what it was intended for (hence the "our take" part). We would view this as an additional trust layer (although a big one we'd argue) in a "know your agent" product. It was mostly intended to show how you can prove an agent of yours created the content seen on a users screen, verify it's integrity, and have this all work in a streaming system.