How much does it matter where your AI model is built?
As AI models become more powerful and embedded into everyday life, many countries increasingly wonder about how important it is to maintain sovereignty over such a foundational technology. On the one hand, AI models are likely to be an increasingly important vector for education, sharing news and culture. On the other, as training costs continue to rise exponentially, only a few actors can afford to build their own foundation model.
In order to explore this topic more, we ran a new survey of 200 respondents in 3 key markets: the US, the UK and India, the host of this week's global AI Impact Summit. For each respondent, we asked them a variety of questions about which countries they were comfortable using models from, how that varied by type of task and which countries they thought had the most powerful models.
In our data, it was clear that respondents did have a geographical preference. Both the US and other G7 geopolitical allies saw a high ratio of comfort to discomfort. By contrast, our respondents were much more likely than not to say they distrusted models from Russia, China, or Israel - and had more doubts over India.
As you might expect, this pattern wasn't uniform by country. Indian respondents were (unsurprisingly) much more likely to trust a model from India - but also much less way of a model from China. In the UK, we saw some wariness of American models, with respondents actually more likely to say they trusted a model from Germany.
Is this just a reflection of underlying sentiment , or is it likely to actually predict behaviour? In order to test this, we asked our respondents across a range of product types whether they would prioritise using the best product possible, or one that came from their own country.
What we saw across all three countries that is that AI chatbots were seen neither as a pure consumer experience where sovereignty didn’t matter, or as an informational product like the news. Instead they were were seen in as somewhere in the middle - around the same level as a search engine.
Another way we looked at this is by asking our respondents how much they trusted the responses of AI models from different countries and varying the use case from the more prosaic, such as shopping, to potentially more sensitive, like news, politics or history.
In general, we saw AI models were across the board less trusted for these more sensitive use cases, and while country of origin did make a difference, it didn’t necessarily exaggerate much the trust gaps that existed already.
Finally, we asked respondents to put in their own words their views on why or why not they would trust a model from another country , and then used an LLM to classify them into different categories. Again, we saw a clear pattern that sovereignty and more to the point, trust, really did matter.
| Theme | Quote | Country |
|---|---|---|
| Data Privacy | I would worry about my data and how they maintain privacy. | 🇺🇸United States |
| If they use or sell my data to large companies | 🇮🇳India | |
| Main concern is Data security and privacy. | 🇮🇳India | |
| Political Distrust | I would feel comfortable using an AI model for any country that isn't leading a genocide or has a very questionable leading strategy. | 🇮🇳India |
| It depends on that country's overall reputation as a force for good in the world. Some countries have historically used technology for nefarious means | 🇬🇧United Kingdom | |
| Cultural Bias | It may be used to exploit and attack on other countries culture slowly slowly. | 🇮🇳India |
| I would not feel comfortable as it is built using a different world view than mine | 🇬🇧United Kingdom | |
| Quality Agnostic | I don't mind where the model comes from, only the answers it gives | 🇬🇧United Kingdom |
| I will choose the quality and accuracy, if another country builds an effective AI, then I will use it, because we should adapt the technologies that changes our lives. | 🇮🇳India | |
| Regulation Needed | If there is transparency in the AI model, I will feel more comfortable in using the AI model. | 🇮🇳India |
| Depends on their background and ethics in the country. | 🇬🇧United Kingdom | |
| I wouldn't know how much trust to place in the AI if I didn't know how it was developed and monitored | 🇬🇧United Kingdom | |
| National Security | I would feel comfortable if I knew that country was an ally of mine and wanted me to succeed in the global arena. I would distrust if it was the opposite. | 🇺🇸United States |
| Depends on the countries allies | 🇬🇧United Kingdom | |
| No Concerns | Yes because they have greater technical capabilities and knowledge. | 🇬🇧United Kingdom |
| I would feel comfortable using an AI model from almost any country as I feel it will be safe | 🇺🇸United States |
Overall, what was clear is that our respondents expected the race for AI leadership to only get closer. While all three countries saw the US as still in the lead today, both the UK and India expected this to have flipped in five years time. That means ensuring we have access to a model that we can trust will only become more important.
