We Said We Would Engage All of Society: Six voices on the AI threshold, and the evidence public authority needs

On September 29, 2026, Bill Gates told Ezra Klein that the artificial intelligence (AI) industry had promised to engage all of society once dangerous capability thresholds were crossed, and that the response had been “complete silence.” This review tests that complaint against selected public remarks by six voices (Jensen Huang, Klein, Barack Obama, Tristan Buckmaster, Geoffrey Hinton and Gates) and three laboratory disclosures. The disclosures do not record a uniform crossing of thresholds: Anthropic withheld general availability of Claude Mythos Preview, its Claude Opus 5.5 system card concludes that the automated AI research and development threshold was not crossed, and OpenAI reports lower chain-of-thought monitorability for GPT-6 Astra. Nor was there literally no response: a voluntary White House accord and the FRONTIER Act, as introduced, both name auditing or verification functions. The narrower finding is that published disclosures let outsiders examine laboratories’ claims but give the public no common, independently reproducible basis for adjudicating the underlying judgments. The exchanges leave distinct questions: channel, instrument, run, provenance, entitlement and jurisdiction. These map onto an observability floor with four dimensions (provenance, configuration, behavioural reproducibility and verifier independence), with access rights and jurisdiction treated as institutional conditions for operating it. Quotations from the World Science Festival interview will be added to §IV in a subsequent version of this preprint after an audio check.