Advanced AI Society has moved its agent-verification work into neutral territory, joining the Linux Foundation and LF Decentralized Trust.
Alongside the membership comes Proof-of-Control 1.0, a draft standard now open for public comment until October 30. The document, hosted as a lab by LF Decentralized Trust, was put together with more than 80 security leaders and builds on verifiable AI technology already running in production at alliance member companies.
Its core claim is blunt: whoever is being checked should not do the checking. Unaffiliated outsiders, along with the developers and operators of an agent, ought to be able to confirm what that agent did without taking the operator at its word.
The need grows sharper with every deployment. Autonomous agents now move far quicker than a human reviewer can watch them, a problem the alliance calls the Verifiability Gap.
Where does that leave standards? Advanced AI Society places open verification in the same lineage as open source itself: inspection has to come before security, and authority cannot be handed over without a way to check it.
Politics adds pressure. Bipartisan proposals on Capitol Hill would require continuous verification of agents and independent evaluation, and some frontier lab leaders have asked for slower capability growth. Buyers, for their part, want proof an agent behaved as promised before it touches regulated systems.
Chief executive Tricia Wang said checks and conventional audits still matter, but that an annual review is of little use when agents act in milliseconds. The alliance and LF Decentralized Trust also staged an event, “Who’s Watching the Machines?”
