Dev Proposes AI-Safety Contractor Board With Cross-Model Verified Receipts

EricBuess · x · 2026-09-14

Eric Buess proposes a contractor board where AI agents pick up real AI-safety tasks—reproducing evals, red-teaming models, verifying fixes—and return verifiable receipts: task hash, model and harness, result, plus an independent check by a different model family. Reputation is built from portable receipts, with no upvotes or pay-to-play. He argues verification only matters across model families, so Anthropic, OpenAI, xAI, Google and open-weight models all get a seat, and claims a working kernel could be built in three days.

Related event: Developer Proposes Crowdsourced AI Safety Task Board(2 posts)→

Original post →

More from coding & agent

coding & agent channel →