The paper separates AI release decisions into two questions: how capable a system may become and who may obtain that capability. It proposes a Multi-Tiered AI Accessibility Model (MAIAM) that sorts access by purpose, safety and capability into three tiers.
Tier 1, “AI for Learning”, is broadly available but deliberately capability‑bounded. Tier 2, “AI for Exploration”, serves research and development under registration, identity verification and increasingly strict usage controls as capability rises. Tier 3, “Full‑Scale Frontier AI”, is accessible only through structured, recorded and accountable channels. The guiding principle is that accessibility should be proportional to capability, risk and safety.
The authors argue that publishing frontier‑class weights forfeits every post‑release safety lever, and cite evidence that safety fine‑tuning can be stripped from open weights for a few hundred dollars, making base‑model evaluations of limited value for derivatives. Criteria for tier assignment cover capability, autonomy, domain consequence and usage/safety axes. MAIAM is combined with the SAFE AI Operating Point to form a two‑part framework: a Capability Boundary that fixes the safety limits not to be crossed, and an Accessibility Boundary that fixes who may obtain a capability below that limit. Capability limits alone do not guarantee safety; a tiered access regime administered by regulators is also required.
Review: The model offers a concrete, governance‑aware roadmap that links technical capability limits with enforceable access controls, addressing a key gap in frontier AI safety.