A Multi-Tiered AI Accessibility Model (MAIAM): Proportionate Access and Capability Limits for Comprehensive Frontier AI Safety

URL SCAN: A Multi-Tiered AI Accessibility Model (MAIAM): Proportionate Access and Capability Limits for Comprehensive Frontier AI Safety
FIRST LINE: # Computer Science > Artificial Intelligence

The Dissection

MAIAM separates capability from access and proposes a gated ladder: bounded mass-market systems, registered research access, and tightly accountable frontier access. Its practical function is not to solve AI’s terminal economic effect. It is to preserve institutional control over the deployment channel while frontier systems become economically decisive.

The paper correctly identifies one narrow fact: releasing weights destroys a developer’s post-release safety levers, and safety fine-tuning is not a reliable barrier if derivatives can strip it. That is a real security observation. It is also the paper’s strongest point—and it is insufficient for the system it claims to govern.

The Core Fallacy

MAIAM confuses control of access with control of the underlying discontinuity.

Even if access is successfully tiered, P1 remains: frontier systems automate cognitive work wherever the cost-performance gap is large enough. The model does not need to be publicly downloadable to displace labor. A small number of controlled owners can deploy it through APIs, agents, internal systems, and firms. Restricting users changes who owns the productive machinery; it does not preserve mass productive participation.

The framework therefore treats a distribution problem as a safety solution. Its Capability Boundary may limit some catastrophic behaviors, but its Accessibility Boundary primarily concentrates the sovereign position. The majority become customers, monitored users, or displaced workers while the operators retain the capital that performs the work.

The second failure is P2. The proposal assumes regulating bodies can maintain a stable, enforceable frontier around capability while incentives favor capability escalation, competitive circumvention, leakage, insider access, jurisdiction shopping, and derivative development. A boundary that depends on durable institutional coordination is not a boundary; it is a temporary administrative preference under competitive pressure.

The paper’s own weight-release argument cuts against its broader confidence. If safeguards can be removed from open weights for a few hundred dollars, then capability-control regimes must defend not only the original system but its copies, derivatives, substitutes, stolen artifacts, and foreign competitors. That is a proliferation problem, not a registration problem.

Hidden Assumptions

• That regulating bodies will possess sufficient authority, competence, and cross-border coordination to administer the tiers.
• That frontier capability can be measured cleanly enough to assign systems to stable categories.
• That capability limits remain meaningful as models are scaffolded, chained, fine-tuned, tool-connected, and combined into agentic systems.
• That access can be restricted without creating overwhelming economic pressure to expand it.
• That identity checks, usage records, and accountability mechanisms meaningfully deter determined actors.
• That centralized control over frontier AI produces broad safety rather than durable ownership concentration.
• That keeping weights closed preserves the relevant safety levers after API deployment, employee movement, leaks, replication, or competitive imitation.
• That the central problem is misuse of capability rather than the lawful substitution of human labor by capability owners.
• That preserving consumption through controlled access is close enough to preserving participation. Under the Discontinuity Thesis, it is not.
• That an administered transition can remain politically stable once productive necessity is concentrated in a small sovereign class.

Social Function

Primary classification: transition management and elite self-exoneration, with a substantial partial truth.

MAIAM is not empty copium. It accurately describes why open-weight release weakens direct safety control. But it converts that narrow security insight into a governance architecture that can make concentration appear equivalent to safety. It offers institutions a respectable vocabulary—proportionality, accountability, boundaries, regulated access—for managing the distribution of dangerous productive power without confronting the death of the mass employment-to-consumption circuit.

Its likely practical role is to legitimate a controlled AI aristocracy: broad access to bounded educational tools, conditional access to increasingly valuable systems, and sovereign control over the frontier. That may reduce some misuse. It does not reverse P3. It organizes the population around dependence on the owners of the systems that replace them.

The Verdict

MAIAM is a credible access-control proposal and an inadequate theory of systemic safety. It can delay uncontrolled proliferation and manage the transition’s surface risks. It cannot stop frontier AI from concentrating productive power, cannot restore mass economic necessity, and cannot defeat the competitive forces that erode durable capability boundaries.

Under the Discontinuity Thesis, this is not a bridge back to the post-WWII order. It is a gatehouse built at the entrance to its carcass.