Open problem 12
The Open-vs-Controlled Access Problem
- current AI
- frontier AI
- AGI
When should model weights, code, data, evaluations, safety methods, and capabilities be open, restricted, delayed, licensed, or classified?
Why it’s foundational
Openness affects science, competition, accountability, misuse, safety research, and geopolitical diffusion. The question is not “open good” versus “closed safe”; both slogans are inadequate.
Why it’s difficult
Open models can democratise capability, enable independent scrutiny, and reduce monopoly power. They can also diffuse dangerous capabilities irreversibly. Closed models may reduce misuse but entrench firms, hide risk, and weaken public accountability.
Hidden assumptions
Open-source advocates often assume diffusion equals democratisation. Closed-model advocates often assume concentration equals safety. Both can be false.
Competing positions
- Open science
- Responsible release
- Tiered access
- Licensing for frontier weights
- Closed APIs
- Government escrow
- Public-interest access for researchers
- Export-control style restrictions
What could make progress
Empirical studies of misuse enabled by open models; measurement of defensive benefits from openness; diffusion models; access-control pilots; legal analysis of licensing model weights; red-team comparisons of open and closed systems.
What it would change
It would shape model-release law, research access, competition policy, export controls, procurement, and safety evaluation.
Sub-agenda
- Which capabilities become irreversible once model weights are released?
- Does openness improve or degrade safety research at the frontier?
- Can tiered access be legitimate and enforceable?
- How should open-weight models be treated under liability law?
- What information should be public even when weights are restricted?
Priority (editor scoring)
Central to diffusion, safety, and competition, but often polarised.
- Importance
- 4/5
- Neglected
- 3/5
- Difficulty
- 4/5
- Actionable
- 4/5
- Robust
- 4/5
- Nat’l+int’l
- 5/5
Where the catalog bears on this
No current catalog instrument resolves this puzzle — which is the point: it is a foundational question the existing rules leave open. Browse the coverage catalog for what the instruments do and don’t say.
Editorial content — a human-authored agenda question, rendered verbatim. No part of this analysis is AI-generated (see the charter).