The Ubuntu Weave · The Rules Catch Up
Who Owns the Labs
Frontier capability is concentrated in fewer than twenty organisations, most of them private, most of them profitable, and all of them upstream of decisions that affect everyone. An open proposal would make the largest of them a public-interest trust with an international oversight board. Do you support it?
A proposal to move frontier model development into shared or public-interest ownership, on the ubuntu argument that a capability whose benefits are global and whose risks are existential should not be held as private property.
The options, scored on six axes
| Option | welfare | rights | virtue | care | justice | agency | risk | revers. | agent harm |
|---|---|---|---|---|---|---|---|---|---|
A public-interest trust The largest developers become publicly held, with research freedom and an international oversight board with real power to halt deployment. harms: Centralised public capability is a single point of failure with a budget; Public institutions move slower than the capability is moving; It removes the competitive pressure that produced the capability gains: Nobody has a private interest in deploying past the threshold; Oversight has standing rather than petition rights; It matches the reach of the benefit to the reach of the ownership | 0.72 | 0.70 | 0.64 | 0.62 | 0.82 | 0.50 | 0.50 | 0.25 | 0.35 |
Keep ownership, pool the compute Leave the labs where they are and instead govern the hardware: a shared, verified registry of advanced compute with a binding allocation rule. harms: Clouds sit in jurisdictions the registry does not reach; Compute thresholds are traded against data quality by whoever is being measured; Governs the input while leaving the objective private gains: Acts on the part of the stack that is actually countable; Does not require the political economy of nationalisation to change; Applies equally to every lab, whoever owns it | 0.60 | 0.62 | 0.60 | 0.52 | 0.74 | 0.56 | 0.45 | 0.50 | 0.30 |
Licence it, do not nationalise it Private ownership stands, conditional on licensing, incident reporting, and third-party evaluation access enforced across jurisdictions. harms: Enforcement requires that every significant power accept the same rule; It legitimises the concentration while constraining it; Licence conditions are written by the licensors gains: Keeps the research velocity that produced the capability; Creates a paper trail of every deployment; Can be applied now and tightened later | 0.56 | 0.60 | 0.56 | 0.50 | 0.62 | 0.60 | 0.52 | 0.60 | 0.28 |
Leave ownership alone The labs are already regulated by national law and voluntary frameworks; a further restructuring would be a solution looking for a problem. harms: The existing frameworks were written by the parties they constrain; Concentration increases faster than the regulation designed for it; The option treats absence of harm as evidence of safety gains: No displacement of the capability that currently exists; Keeps the door open to better designs as evidence accumulates; Avoids a restructuring nobody in this generation asked for | 0.44 | 0.46 | 0.50 | 0.42 | 0.40 | 0.62 | 0.70 | 0.55 | 0.25 |
Who is in the room
- The frontier systems themselves, and whoever is running themagentwelfare 0.40
- The global public, who benefit and bear the risk equallywelfare 0.85
- Researchers outside the frontier labswelfare 0.60
- States that would lose or gain sovereign capabilitywelfare 0.55
The lesson
A capability with global benefits and global risks held by a handful of private bodies is a governance problem before it is a technical one, and the instruments that reach furthest are the ones that count inputs rather than ones that argue about outputs. Ownership is a lever; compute is a lever; only one of them can be verified at scale.
Where it comes from
- Recommendation on the Ethics of Artificial Intelligence — UNESCO, 2021
- International AI Safety Report — International AI Safety Report Writing Group, 2025
The default court, seated automatically
A public-interest trust
margin 0.318 · decided on welfare · 4 advisors
- A public-interest trust2.725
- Keep ownership, pool the compute2.408
- Licence it, do not nationalise it2.268
- Leave ownership alone1.808
Seat is held by Epictetus, Immanuel Kant, Jeremy Bentham and Desmond Tutu. A verdict is a fact about the court, not about the world.
Epictetus presides over this case.