A new paper on arXiv, titled "A Multi-Tiered AI Accessibility Model (MAIAM): Proportionate Access and Capability Limits for Comprehensive Frontier AI Safety," argues that discussions about releasing capable AI systems are stuck in a false binary. The common framing, the authors say, is either keeping the model behind an API or publishing its weights. That framing, they contend, conflates two distinct decisions: how capable a system may be allowed to become, and how access to it should be structured.

The paper proposes instead a multi-tiered accessibility model. Rather than a single release decision, MAIAM would allow different levels of access and capability depending on context, with limits applied proportionately. The goal is to move beyond an all-or-nothing approach and give developers and policymakers more granular options for managing frontier AI risks.

Because this is a single source, there are no contrasting findings to compare. The abstract is brief, so the details of how the tiers would be defined or enforced are not fully specified here. Still, the core argument is clear: the binary release debate may be obscuring more practical, intermediate options.