Craig Horton
The Transformation Brief
Good morning. This week Nvidia, Microsoft, Meta, IBM, Dell, Palantir and Hugging Face put their names to a call for the United States to back open-weight AI. The argument is about competition, national control, and letting organisations download, modify and run models on their own infrastructure.
The question I keep coming back to is what happens when the supplier can no longer carry the responsibility for you.
The lead signal
The model is open, the accountability is yours
The American AI industry has moved towards open-weight models, pushed by Chinese models closing the performance gap and by concern that a few closed providers could control access to the technology. The industry letter says an open ecosystem will spread capability, reduce dependency and let organisations control their own data, models and infrastructure.
Open Weights and American AI Leadership, 25 signatories, 24th July 2026
There is a lot to like in that. An organisation can run the model where it chooses, inspect how it behaves, adapt it to its own work, and stop tying every decision to one provider's commercial terms.
Worth noting who did not sign. OpenAI, Anthropic and Google, the companies selling closed frontier models, stayed off the letter.
Open weight is not the same as open. The weights may be published while the training data, source code and development decisions stay closed, and once those weights are out the provider cannot recall them when something goes wrong. The same freedom that lets an enterprise control the model makes monitoring, withdrawal and enforcement harder.
That changes the buying decision. I would want to know who owns the model after it has been modified, who tests it against new threats, who monitors its decisions, who reads the licence, who responds to an incident, and who has the authority to turn it off.
Each answer needs a name, a budget and a place in the operating model. Without those, you have downloaded capability and inherited a liability you have not designed yourself to manage.
One-question pulse
This month we are measuring the Strategic Alignment driver of the Blueprint. Your answer joins the Blueprint Index, which we publish in the next issue.
Can your board or executive team articulate your organisation's AI strategy in one sentence that would survive a 30-second explanation to a new hire?
From my office
I have had to draw the same line with the AI Business Value Framework. The framework is my intellectual property, the connector code is MIT licensed, so anybody can inspect it, test it and run it.
That makes the software easier to challenge and to distribute, while every verdict still carries its rules and its audit trail. Open code does not make the decision ownerless, somebody still has to accept the evidence, act on the verdict and carry the result.
The same holds for open-weight models at enterprise scale, technical access can be spread widely, authority cannot be left undefined.
The question I would take into the room
Before you approve an open-weight model, can you name the person who owns its security, evaluation, licence, operating cost, incident response and withdrawal decision once it is inside your organisation?