France’s government is using DeepSeek for coding through a shared AI gateway, according to Takahiro Anno’s account of interviews with French officials and researchers. The choice puts open-weight models and control over infrastructure ahead of access to the very latest closed models for some everyday government work.
DeepSeek Watch analysis
What happened
In a report based on 26 interviews in France, Anno says the French Digital Agency operates an AI gateway for government departments. Its interface is compatible with OpenAI’s API and its code is open source, but it cannot access closed models from US companies. Instead, it runs open-weight models on cloud infrastructure certified in France.
DeepSeek was described to Anno as the gateway’s most-used model, mainly for coding; Alibaba’s Qwen was also reportedly used frequently. Anno says officials cited sovereignty as a reason for this approach. The report also says France’s AI safety agency, INESIA, is checking risks, while interviewees said there was no research showing Chinese models exhibiting the feared behaviour they were investigating.
Why it matters
This is a concrete example of AI sovereignty shaping procurement and deployment, not just appearing in a policy speech. A shared gateway gives departments a common route to models, while the reported restrictions and locally hosted open weights give France more control over which systems it can use and where they run.
The trade-off is explicit: officials told Anno that being a year behind frontier models was acceptable for routine administrative work, but cybersecurity was an exception where near-frontier performance mattered. That is a more useful distinction than the usual assumption that every public-service task needs the newest model.
Our read
France’s approach is a practical test of whether governments can accept less than the frontier in exchange for greater control. The notable detail is that DeepSeek is reportedly being used in real government work, but the account rests on interviews and does not establish how widely it is deployed or how it performs. Sovereignty is a compelling aim; the engineering and oversight still have to earn it.
What to watch
- Whether France publishes further details on the gateway’s models, safeguards and uptake.
- Whether other European governments adopt locally hosted open-weight models for routine work.
- How INESIA’s evaluations inform which models can be used, especially for sensitive tasks.
Discussion spark: For routine government work, is control over where models run worth accepting a year’s lag behind the frontier, or should public services prioritise access to the strongest available models?
Sources and evidence
not affiliated with or endorsed by DeepSeek