Category: Ownership / voluntary · Horizon: Short · Consensus: Niche · Feasibility: Uncertain
What it is
A windfall clause is a promise made in advance by AI companies: if they ever earn extraordinarily large profits from advanced AI, they will give a significant share away for the common good. The commitment costs little today, because it only applies if a company becomes enormously dominant, but it would matter a great deal if that happened.
Why it matters in the age of AI
If one or a few companies build AI capable of doing most work, they could capture a share of the world’s wealth with no precedent in history. A windfall clause is an insurance policy against that scenario, agreed now while no one knows who the winner will be, which makes it easier to sign up to.
Evidence and examples
- The original proposal. The idea was first suggested by philosopher Nick Bostrom in 2014 and developed in a 2020 report by Oxford’s Centre for the Governance of AI. Its authors defined windfall profits as a company’s profits exceeding about 1% of the world economy. As an illustration, they suggested firms give away a rising share of profits on a sliding scale, from 1% at the lowest threshold up to half at the highest. Several of the authors now hold senior governance roles at leading AI labs.
- No company has signed one. The clause remains a proposal, and the report did not settle where the money should go.
- Related real-world structures. In October 2025 OpenAI restructured so that its non-profit, the OpenAI Foundation, holds a 26% stake in the company, worth about $130 billion at the time, with an initial commitment of $25 billion to health and AI safety. Several AI companies, including OpenAI and Anthropic, are now public benefit corporations, legally required to weigh a public mission alongside profit. Neither is a windfall clause, but both point in a similar direction.
Arguments for
- Can be adopted quickly, without waiting for legislation.
- Cheap for companies to agree to now, since it only bites in an extreme scenario.
- Works across borders, unlike national taxes.
- Signals publicly that the gains from AI should be shared.
Arguments against
- Voluntary promises may be broken, renegotiated or legally challenged by shareholders.
- The threshold may never be reached, even if AI causes widespread job losses through many smaller companies.
- Leaves decisions about how to share the money with private companies, not democratic institutions.
- Charity is no substitute for rights: people may prefer a guaranteed share to a donation.
Open questions
- Who should receive the money: a global fund, national governments, or citizens directly?
- Could the clause pay into a citizens’ wealth fund rather than charity?
- Should governments make such clauses a condition for building the most powerful AI systems?
Leave a Reply