Preston Byrne argues embedded AI evaluators will end up where GARM and Ofcom did
In 'Who Aligns the Aligners?' Byrne makes the case that Amodei's embedded-evaluator and coordinated-standards design will be captured for political censorship, drawing the precedent from GARM, which he says 'worked with global regulators to pressure companies like Twitter' by wielding collective advertiser power. He cites the Ofcom/SaSu case, where the regulator accepted a geoblock as adequate remediation and then reversed after 'circumventing the geoblock using VPNs' under political pressure, as the pattern AI evaluation will follow. His structural claim is simply that 'regulators are subject to political control' and will therefore do political things, and that defection by non-compliant actors makes enforcement selective rather than universal.
↳ Follow the thread