Observed 9 September 2026. Scope: Fathom public strategy, workshop reports and evaluation-industry interviews. The California instrument is separately assessed in fathom-bill.md.
Fathom explicitly seeks legislation to create demand for an independent AI-verification market. Its model combines public outcome requirements, accredited private evaluation and legal/commercial benefits for passing products. Its philanthropy-only funding description is self-report, not a completed donor audit. Source: http[local research file] (substantive approach read).
The 16 March Paris workshop account describes a February 2026 IASEAI event and Gillian Hadfield keynote. Its published full report contains a simulation, not an executed regulatory arrangement. PDF pages 9-11 identify unresolved revocation/appeal rights, transparency versus proprietary methods, and incentives when developers pay evaluators. Participants explicitly compare this to credit-rating failures: limited verifier competition can make revocation difficult; government, evaluator and developer can shift responsibility between themselves. The report also records participants who thought the structure improved responses to harm. These are organizer-reported simulation findings, not evidence of an actual cover-up, crisis operation or captured regulator. Sources: http[local research file] and http[local research file] . Landing page substantive body read; PDF targeted pages 9-11 plus scoped extraction of other sections. Original retained with hash in originals-manifest.json; extraction contains spaced lettering.
Ashby2026 connects voluntary verification to insurance pricing, coverage and liability, explicitly describing these as incentives for uptake. Its children workshop proposes an independent AI-rating institution with access to proprietary developer data; it acknowledges unresolved age-verification/First Amendment tensions. Its separate fraud workshop advocates identity verification and cross-sector data exchange at high-risk interactions. These are workshop proposals, not established institutions or a single operational identity system. Read PDF pages 8-10 and relevant extracted sections. Sources: http[local research file] and http[local research file] . Original retained, hash in manifest. Do not infer participants endorsed every proposal or reproduce unsourced factual claims inside the report.
The 21 July evaluator article interviews Rumman Chowdhury of Humane Intelligence PBC, Shea Brown of BABL AI and Patrick Sullivan of A-LIGN. It identifies test design, access rights, commercial relationships and comparability as unresolved practical issues, advocating professional standards, accreditation, independent oversight and liability insurance. Interviewed aspirations are not adopted industry rules. The PBC must not be silently merged with the similarly named nonprofit on a partnership roster. Source: http[local research file] (full substantive article read; landing http[local research file] is same source family).
Whole-case implication
There is a concrete economic mechanism worth investigating: a formally optional certificate can become commercially difficult to avoid if insurers, procurement officers or litigants rely on it. This is a supported design intention and an open implementation question. Actual policy endorsements, procurement clauses, insurer schedules and declined-coverage decisions would establish whether it operates, by whose standards and with what contestability. We have not acquired those instruments in this run. Public-interest conviction, market-building and an institution's desire for a role can coexist. None alone demonstrates illicit control.