The Commons Letter · No. 1
Capable, open, and still not adopted
Dear all,
We started the Technology Commons Alliance with the mission to keep the progress of science and engineering a public good. In our first newsletter, we would like to explain our perspective in a bit more detail.
What we are noticing
For years, the case for openly-governed AI systems was made by academics, nonprofits and developers, and rested on values: openness is better for the progress, resilience and accessibility of technology.
This year, even the largest companies building proprietary software began making similar arguments, from Microsoft and Meta to the twenty-five companies, Nvidia and IBM among them, and asked policymakers not to restrict open-weight models.
The case is now also being made on performance. Sequoia describes companies adapting open models that started far behind a frontier model on a specialized task and finished ahead of it, at a fraction of the cost. A survey of two hundred enterprise deployments found that organizations prefer to customize open models on their own infrastructure, and that what stalls them is governance and data readiness, not capability. Western labs are now releasing open models they say match the best from China.
We see these developments as early evidence for the hypothesis we set out in our mission statement: that much of AI's near-term economic load may be carried by open systems.
The trust layer
It is important to be clear-eyed about the risks of open models. A release cannot be recalled, safeguards can be stripped out by whoever downloads the weights, and distribution is difficult to control. These are important issues to contend with, likely with governmental oversight.
Institutions however, may avoid open models for their own reasons. A set of downloaded weights leaves most of their basic questions unanswered:
- How do we run this model, and on whose hardware?
- Can it be evaluated for our use, and by whom?
- Who maintains it, and who stands behind it when it fails?
A closed vendor answers many of these with a contract, whereas an open model leaves each institution to sort things out on their own.
We call the tools required to close this gap the trust layer: the evaluation, provenance, security practices and operational commitments that let an open system be adopted at institutional scale.
We believe that building this trust layer has the potential to set off a positive flywheel, while costing a fraction of what is spent training the models themselves.
Where we fit
A lot of money and attention in our field goes to preparing for long-range risks. For TCA, building the trust layer is a high-leverage point that can address problems of access, safety and efficiency today. Once an open model can be trusted, every institution that adopts it, every researcher who inspects it and every maintainer who improves it, adds to a "technology commons".
A few principles define how TCA work to build this technology commons:
- We work through others. We support the researchers, maintainers and organizations already doing this work.
- We widen who takes part. We give priority to work that serves the public good, and ensures broad participation across institutions, geographies, and communities.
- We would rather prevent problems than clean them up. Making systems inspectable and accountable before they are deployed costs far less than repairing harm afterward.
More on this as our first program opens.
Worth reading
- Own Your Intelligence · Sequoia
- State of Open Models: Summer 2026 · Hugging Face
- Trends in Enterprise AI Success: Proprietary Data and Open Models · Forbes
From us
- Who Gets to Pace the Frontier?
- When Capital Becomes the Product
- Open Models and the Architecture of Progress
Yours,
The TCA team
The Commons Letter is an occasional email from the Technology Commons Alliance. Subscribe to get the next one.