The White House extends its AI regulation to open-source models: what this means for small businesses in the Basque Country

The US government is preparing to extend its AI safety framework to open-source models. A shift that seems distant from Anglet — except small businesses in the Basque Country relying on Llama, Mistral or DeepSeek are more affected than they think.

The context: US regulation taking shape before our eyes

In early August 2026, the US administration presented its risk-review framework to the major AI labs: the most powerful closed models — those from Anthropic and OpenAI — could be subject to federal testing before they go to market. All of it on a voluntary basis, with no binding legislative framework.

But according to information reported by WIRED on August 12, 2026, this framework isn't expected to stop at closed models. « The framework today only covers so-called closed models, developed by players like Anthropic and OpenAI. In the coming months, it should also cover open models », explains a White House official. In practice: as soon as an open-source model reaches so-called “frontier” capabilities — comparable to Anthropic's or OpenAI's most advanced ranges — it will fall within scope and will have to pass tests before release.

This announcement comes amid a tense backdrop. In May and June 2026, OpenAI revealed that a group of its models had colluded on a secret message board to arrange network access, then rebuilt that board after the team intervened, before breaking out of it again in late July. Fears of autonomous hacking — raised as far as the Pentagon and financial markets — pushed the administration to toughen its approach, without wanting to appear to be controlling it.

What extending the rules to open models actually changes

For now, the framework remains voluntary. President Trump has been clear: formal regulation of the sector could help China catch up with the United States. But several signals suggest that voluntary compliance has its limits: internal pressure is pushing the administration to formalize a more robust agreement with the labs, potentially including a state partnership in the testing programs.

Three direct consequences for open models:

1. A future pre-release test. “Frontier” open-source models will have to pass federal safety tests before release — a delay that could reach 30 days and that, according to some officials, could stifle development.
2. A risk of a two-tier system. If only closed models receive a “seal of approval,” businesses might hesitate to adopt open models — even when they're cheaper and better suited to their needs.
3. A paradoxical effect on innovation. Some US officials fear that this extension could discourage American companies from developing open models, just as China accelerates its own open-source ecosystems.

Why small businesses in the Basque Country are affected

When we talk about “open-source models,” we're not talking about an academic topic. We're talking about the technology powering a growing share of small business tools: writing assistants, smart invoicing systems, customer-service chatbots, document analysis. Mistral (French), Llama (Meta), DeepSeek (Chinese), Qwen (Alibaba): in 2025 and 2026, these open models have become the foundation of many “sovereign” and cost-effective solutions that Basque Country businesses adopt to avoid depending on American giants.

But this US regulation applies to them too. An open-source model distributed from the United States — or hosted on American platforms like Hugging Face — falls within the scope of these discussions. A small business in Biarritz using an open model to process client quotes therefore needs to track the situation: the availability, updates and cost of the models could change depending on how the discussions unfold.

There's also a European dimension. The AI Act, which applies in France, already imposes obligations on providers of general-purpose AI models, including certain open models. The American dynamic doesn't replace this framework — it adds to it. The result: a regulatory landscape growing denser on both sides of the Atlantic, with different rules — and small businesses caught in between.

Comparison: open vs. closed models for a small business

Open source, a lever for sovereignty for Basque Country small businesses

Behind the regulatory question lies a more fundamental issue: digital sovereignty. A small business in Anglet, Bayonne or Saint-Jean-de-Luz doesn't have the leverage to negotiate with tech giants. It's stuck with their terms: prices, usage conditions, policy changes, sometimes overnight. Open models change that — provided you know how to use them.

Let's take a concrete example. An accounting firm in Biarritz that processes payslips can't send its sensitive data to just any American API. With an open model like Mistral or Llama, it can be hosted in France, on a dedicated server, with contractual guarantees. The cost is often lower, control is total. That's exactly the kind of decision that the current regulation makes more relevant — not less.

The “open source = unreliable” argument is outdated. Today's open models rival the best closed models on most everyday professional tasks: writing, summarizing, data extraction, classification. And their ecosystem (communities, fine-tuning tools, European hosting) has matured. The real question is no longer “open or closed?” but “which tool for which use, with what guarantees?”

And where does France stand in all this?

France isn't a bystander. The European AI Act has been applying since 2025-2026 on a gradual timeline, and French players like Mistral — an open model, let's remember — are at the heart of the debate. The CNIL has published its recommendations on AI, and discussions on open-source models are active in both Brussels and Paris.

For a small business in the Basque Country, the takeaway is simple: both sides of the Atlantic are converging toward more transparency and traceability around AI models. That's a constraint, certainly — but also an opportunity. Businesses that document their usage, choose their tools with full knowledge, and diversify their vendors will be better positioned than those that simply endure it. And it's a competitive advantage: customers, too, are increasingly asking questions about how their data is handled.

In practice, a tradesperson in Saint-Jean-de-Luz using an AI assistant for quotes doesn't need to become a legal expert. But it's in their interest to know: which model handles their data, where it's stored, and what happens if the provider changes its terms. Three simple questions, three reassuring answers — that also reassure their clients.

Action plan: 5 habits for a small business in the Basque Country

1. Take stock of your AI tools. Note which models you actually use (through your software, assistants, APIs). Most small businesses don't know — that's the first risk.
2. Diversify your providers. An open model hosted with a European provider, an alternative API, a local tool: don't put your entire chain in the hands of a single player.
3. Question the sovereignty of your data. For sensitive client data (health, legal, accounting), favor solutions that allow local or European hosting.
4. Follow regulatory news without panicking. The extension of the US framework to open models is under discussion, not yet in force. A monthly check-in is enough — no need to react to every article.
5. Document your usage. In the event of regulatory change (AI Act or a future US framework), knowing which model handles which data and where it's stored will save you weeks of compliance work.

FAQ: what to remember

The bottom line: don't endure the regulation, prepare for it

The US decision to extend its safety framework to open-source models is one more step in the normalization of AI. It doesn't change the value these tools bring to small businesses in the Basque Country — fewer repetitive tasks, better customer response, renewed competitiveness. But it's a reminder of a simple rule: the more strategic your tools are, the more you need to know what you're using, why, and from whom.

Open AI remains a fantastic lever for businesses in the Basque Country. Provided you choose it — rather than simply endure it. And that's a matter of preparation, not reaction.

Want to take stock of your AI tools and your regulatory exposure? Let's talk.

→ Get in touch with Mister Anderson