The question layer
What the fight is about
Questions are the stable spine of the register. Beneath each one sit the asks: the specific outcomes actors have requested, with who is for, who is against, and how it is going.
Should state or federal law govern frontier-model safety?
The central jurisdictional fight in US AI policy. One side asks Washington to preempt or pause state AI statutes, by moratorium, executive action, or a federal framework that displaces state law. The other side asks states to keep legislating in the absence of federal rules, and asks Congress not to strip that authority without replacing it.
- Wants fewer rulesPreempt or pause state frontier-safety statutes2 for · 3 againstPartially adopted
- Wants more rulesEnact one federal frontier-transparency framework2 for · 2 againstActive
- Wants fewer rulesCreate a federal AI regulatory sandbox waiving federal rules (SANDBOX Act)1 for · 0 againstActive
- Wants fewer rulesDeem federal or EU-code compliance sufficient under state AI laws1 for · 0 againstActive
Who bears liability when a deployed AI system causes harm?
Whether responsibility for harms attaches to the developer of a model, the deployer who put it in front of users, or nobody without proof of fault, and whether liability relief should be traded for participation in government safety programs.
May developers train on copyrighted works without a license?
Whether training on lawfully accessed copyrighted material is fair use (US), a text-and-data-mining exception with or without opt-out (UK/EU), or an act requiring licenses and remuneration, and what developers must disclose about what they trained on.
When should high-risk AI obligations start to bind?
The EU AI Act set application dates; industry asked to postpone them, civil society asked to hold the line. The question covers 'stop the clock' asks, transition periods, and the Digital Omnibus, which moved high-risk obligations to December 2027 (Annex III) and August 2028 (Annex I).
What must frontier developers disclose — and to whom?
Published safety frameworks, incident reporting, model cards, whistleblower channels: which disclosures should be mandatory, at what capability threshold, enforced by whom. The live fight behind California SB 53, the NY RAISE Act, and the EU Code of Practice.
- Wants more rulesMandate published safety frameworks and incident reporting6 for · 4 againstPartially adopted
- Wants more rulesMandatory pre-deployment government review of frontier models1 for · 1 againstActive
- Wants fewer rulesLimit AI Act foundation-model obligations to a transparency baseline1 for · 0 againstRejected
Should open-weight model release be restricted?
Whether governments should restrict, monitor, or affirmatively encourage the public release of model weights, the question NTIA answered with 'monitor, don't restrict' in 2024 and the AI Action Plan answered with 'encourage' in 2025.
Should access to advanced AI compute be export-controlled?
Chip export controls as AI policy: the Diffusion Rule and its rescission, chip-smuggling enforcement, location verification, and whether compute governance is a national-security necessity or a self-inflicted market loss.
What say do workers have over AI deployment decisions?
Whether law should require notice, consultation, or bargaining before AI systems replace or manage workers, and whether a worker's voice, likeness, and output can be simulated without consent and compensation.
Should states impose high-risk AI duties on consequential decisions?
Whether states should impose developer/deployer duties, impact assessments, an anti-discrimination duty of care, consumer notice and appeal, on AI used in consequential decisions, and whether such comprehensive acts survive industry pushback and federal challenge. The core fight of 2024–26: Colorado SB 24-205 (delayed, enforcement suspended, repealed and reenacted as a transparency-only framework), Texas TRAIGA (pared from the Colorado model to intent-based prohibitions), Virginia HB 2094 (vetoed), Illinois HB 3773 (employment-only, in force).
Should the UK legislate binding frontier AI rules?
Whether the UK should move from its 2023 non-statutory, regulator-led framework and voluntary frontier commitments (Bletchley/Seoul, AISI testing) to binding legislation on developers of the most powerful models. A bill was promised in 2024, repeatedly delayed through 2025, and absent from the May 2026 King's Speech, with AI instead threaded through sectoral bills and sandbox powers.