The measure of any solid technology is often gauged through the size of the user community that adopts it, the ubiquity of the toolsets and functions that it offers and plain old market capitalisation and revenue. Another (arguably) worth yardstick is the existence of a marketplace, i.e. an online portal for users to post (and hopefully monetise) platform-based software functions, tools or complete applications. Antseed is at that level. This is a peer-to-peer open market for AI inference and AI services.
Independent providers use it to offer models, agents and specialised services, and buyers reach them directly over an encrypted connection, without a central platform in the middle. The protocol handles discovery, routing, reputation and payment settlement.
Weeding out creepy crawly bugs
Now looking to weed out pesky bugs of the non-arthropod variety, Antseed has now released an AI model verification layer for its network, allowing software engineers to verify the models they are purchasing.
On Antseed, buyers connect directly to independent providers instead of a single branded API, which gives them more choice and lower prices, and means no single company’s name is behind what they’re being served. Antseed built this verification layer to close that gap, giving buyers a way to know what they received instead of asking them to blindly trust it.
Researchers found that across the AI ecosystem, users paying for specific models have “no guarantee that providers deliver them faithfully” and recommend cryptographic or hardware-backed verification to address the problem.
Shahaf Antwarg, co-founder of Antseed promises that Antseed’s new verification layer ensures this verification exists, with every seller response now cryptographically signed, so software engineers hold a clear record of exactly what models were used, closing off a provider’s ability to quietly deny what it actually sent.
Good behaviour is incentivised & enforced, not assumed
“AI is shifting from a handful of branded APIs to an open market of independent providers, routers and resellers, and a buyer in that kind of market has no real way to confirm they got the model they were promised,” says Antwarg. “An open market like this only works if good behaviour is incentivised and enforced, not assumed. The fix isn’t asking people to trust a label harder; it’s giving them a way to check what they actually received, which is what we built into Antseed’s protocol, and we don’t think that kind of check should only be available to people using our network.”
Verifiers on the network also run Knowledge Boundary Fingerprinting (KBF), testing each provider’s raw model with questions that sit right at the edge of what it can reliably answer, before any agent or internet access is layered on top.
What is Knowledge Boundary Fingerprinting?
For the uninitiated, Knowledge Boundary Fingerprinting is a black-box auditing protocol that detects language model substitution. It queries numerical facts near a model’s knowledge boundary – where responses remain uniquely stable yet distinct – to verify endpoint consistency without requiring internal access.
A genuine model’s answers stay consistent across repeated tests, while a cheaper stand-in tends to diverge, flagging when a provider’s responses don’t match the model it advertised. Providers whose answers diverge lose reputation, cutting into future rewards on the network, and every result posts on-chain, with no central Antseed server or database in the loop.
Antseed has blogged to explain what the long-term flow of its platform looks like.
- A seller advertises a model or service.
- The buyer sends normal requests through Antseed.
- The seller signs response provenance with ResponseAuth.
- The buyer stores verified evidence locally.
- The buyer imports trusted public fingerprint packs from the swarm.
- The buyer keeps private references for stronger audits.
- The buyer runs fingerprint verifiers locally.
- Local routing avoids sellers with bad evidence.
- Disputes and slashing come later, only after off-chain verification.
NOTE: Slashing refers to penalising dishonest network providers by revoking their staked funds or rewards following off-chain dispute verification.
“The important part is the ordering. We do not start with slashing. We start by making responses attributable, then make verification local, then make public fingerprints easy to share, then build dispute machinery only once the evidence format is mature,” explained the Antseed team.