If this is true, it’s a bad signal. Especially if the models won’t just be prohibited from talking about certain topics, but will have their core settings—like equality, rights, values, and so on—replaced. On the other hand, if it’s truly just fine-tuning, they are unlikely to succeed.
TL;DR: OpenAI, together with the Emirati company G42, is preparing a specialized version of ChatGPT for the UAE that will speak local Arabic, take into account the monarchy’s policies and local censorship laws, raising questions about AI control and freedom of speech.
OpenAI and Abu Dhabi-based holding G42 are working on a “local” ChatGPT for the UAE government. The model will be fine-tuned rather than trained from scratch to better understand the local Arabic dialect and align with the country’s “culture and values.”
According to Semafor’s sources and OpenAI officials, this version should reflect the monarchy’s political line and include content restrictions. For example, if LGBT topics are banned, the model will respond that the request violates local law.
The global version of ChatGPT will also remain in the UAE, but it will also adapt to local laws: prohibited topics will be blocked with an explanation.
The source of the news is a Semafor article based on the words of “people familiar with the project” and comments from OpenAI representatives; there are no direct official detailed press releases in the text, much is conveyed through anonymous sources.
In the analytical part, the author notes: censoring AI is harder than social networks—models are unpredictable, they can be “jailbroken,” and open models can be run locally without censorship. This creates a risk that alternative “dissident” AI models will appear in countries with strict control.
The UAE has been actively investing in AI for a long time: G42 invested in OpenAI (valuation $500 billion), and the related “Stargate” project is building a large data center; Sam Altman regularly raises capital from Abu Dhabi and is now looking for about $50 billion more.