The Answerable Commons: Governing the Epistemic Standards of AI Intermediaries

General-purpose answering systems now settle, for very large publics, which sources count, how answers are framed and what is left out. Their standards answer to no forum in which they can be contested as such. A companion paper argues that this can enclose the governance of knowledge, displacing a community’s capacity to sustain and contest the practices on which it depends. This paper asks what the law can see of that problem and what would answer it. It grounds duties of answerability in consequential epistemic intermediation, not in enclosure, and argues that binding remedies against a private standard-setter are warranted where it performs a delegated function, public or part of an institution’s authority over persons, on which those affected depend. An audit of UK, EU, US and Chinese law as at October 2026 finds that existing regimes reach acquisition, copying, corpus disclosure, the fact of generation and some selection standards, but provide no forum in which the epistemic standards of general-purpose answering systems can be contested as such. The paper proposes an answerable commons: provenance duties; review of the reasons for standards before an independent forum with power to disapply an inadequately justified standard in designated deployments; participatory setting of the forum’s framework; a share for those affected in standards for designated public functions; rules that keep automated assessments from becoming verdicts; obligations to sustain public epistemic capacity that substitution materially impairs; and a statutory conflict rule and duty of candour. It states what this architecture cannot do.