Three days after launch, Claude Fable 5 and Claude Mythos 5, two of Anthropic's newest models, were switched off for almost three weeks. Not a bug. A US export-control decision. Few teams had built anything on them yet but think about what would happen if your team worked with that tool and it just stopped.

Daniel Naeff (ETH AI Center) made exactly that point on Monday at the Zurich AI Festival: access to AI can be cut by decisions you have ZERO say in. Rules change, providers change terms, a tool gets suspended, and you dont control any of it.

When I ran a governance consistency review for an international NGO in Geneva, six interdependent governance documents and 220 inconsistencies, I used Claude heavily. But I started with one small part, then applied the method to the next and only scaled up once it held, adding guardrails and checkpoints as the project grew. Every source file and output lived in my own files, not in a chat history. Claude did a lot of the work. It wasnt holding the project.

The speakers on Monday called it the sandbox approach: test, observe, learn, scale. It works for a 10-person team as well as for a country. At team level, it looks like this:
- Test on one real task before rolling out to everyone
- Observe where the tool helps, and where it quietly gets things wrong
- Keep your sources and outputs somewhere the tool doesnt own
- Scale only what survived the test

Skip that, and the cost shows up later: licenses nobody uses, rework on outputs nobody checked, or a process that stops the day a provider changes its terms. For Swiss SMEs, theres already a version of this with support included: the Canton of Zurich runs an Innovation Sandbox for AI, with regulatory guidance for pilots. This is the part I do with teams: pick the first real task, test it properly, and set things up so the team keeps control, not the tool.

Training, a workflow with or without AI, or something more technical, happy to talk if youre at that first step.

PS: on Monday, one of the speakers asked 200 people who had read the Sandbox Manifesto. My hand stayed down too 😄Here are the links:

https://ai-startup-center.com/the-sandbox-manifesto/

https://www.zh.ch/en/wirtschaft-arbeit/wirtschaftsstandort/innovation-sandbox.html

https://www.anthropic.com/news/fable-mythos-access

PS2: You are still on time for the Zurich AI Festival!, See the program here.