AI News

Skin is a proposed profile and record layer intended to help AI agents and language models use facts published and controlled by a person or company. The project addresses a problem in which models describe organizations from stale training data, scraped pages, or guesswork, leaving owners without a correction channel and agents without a reliable way to confirm a source. A publisher would verify ownership through a DNS TXT record or OAuth sign-in, describe the organization in plain conversation, and have Skin structure the information into machine-readable fields. The record would be indexed for retrieval and queried on demand, so agents could receive the owner-confirmed version currently available rather than a frozen copy from a model’s training cutoff. Skin says the format is designed to support native rendering in different languages and does not require publishers to build a machine-readable website, maintain an endpoint, or synchronize a file. The service is intended to complement search and retrieval-augmented generation: an agent would find candidates through its normal search or browsing process, then query Skin to verify identity and retrieve current facts. The proposal is positioned as an alternative in scope to llms.txt because it adds ownership verification, an index, and an on-demand query interface rather than relying on a static file that agents must discover and trust. Skin acknowledges that verification proves who published a record, not that every claim is objectively true. The project is currently a landing page validating demand, and it says LLM providers do not yet query Skin. Publishers may eventually pay a subscription, while querying is planned to be free for agents and developers; early users are invited to join the waitlist.