Why the Swedish government blocked ChatGPT but chose to keep Microsoft Copilot

Sweden has blocked government employees from accessing Anthropic’s Claude and OpenAI’s ChatGPT, citing security concerns. However, government workers can still use Microsoft Copilot, which is built on OpenAI’s technology.

Both Copilot and ChatGPT share the same “brain,” developed by OpenAI, which lays the foundation for the AI.

While Copilot is separate from ChatGPT – as it functions as a generative AI assistant built within the Microsoft ecosystem – it is built on many of OpenAI’s advanced AI models, which also allows ChatGPT and Claude integration.

Why is Sweden keeping Copilot?

Sweden is ditching ChatGPT and Claude in favor of Copilot, as the tool has been reviewed by the government and strict guardrails have been put in place, turning it from a commercial product into an AI tool specifically designed for government operations.

One X user explained the real reason for the switch: Sweden doesn’t have control over what the AI tool sees and logs.

The digitalization director, Mikael Wettercrantz, who oversees digital strategy and technological advancements in Sweden, suggested that using unregulated AI in government could negatively affect Sweden’s information security.

In this context, unregulated AI models aren’t checked or given the OK by the government, which could have negative consequences if a data leak were to occur.

When people use ChatGPT or Claude, their prompts are sent to the company’s external servers, and the company then decides how that data is handled, which is a concern for those working with confidential information.

ChatGPT and Claude are dangerous, declares Sweden

Copilot passed security checks by the Swedish government, and its safety features were enhanced even more after that.

However, ChatGPT and Claude were found to be problematic after an investigation by a Swedish government institution, which identified issues with input and information handling.

The government also didn’t like the kind of sensitive information that was shared with the chatbot.

While this might not be the fault of OpenAI or Anthropic, it raises concerns surrounding a lack of native technology and overreliance on US models.

We rely on US tech, and Europe’s just figuring that out

The majority of technology used in Europe is developed by the US, a country with different laws and regulations surrounding information security and data handling.

US dominance in Europe has made developing EU-sovereign technology seem less urgent or perhaps undesirable.

However, AI ethics and digital sovereignty expert Pernille Tranberg explained that this is a problem, especially as European countries discover the inherent dangers of certain US technologies.

“We are making the same mistakes we did with Google, Microsoft, and Facebook. Using US tech and abandoning our own services in Europe. Now, we are trying to get rid of all 3, but then we are building dependencies on OpenAI and Anthropic,” Tranberg said.

Government workers can still use Claude and ChatGPT

While Sweden has blocked browser versions of ChatGPT and Claude, government employees can still access both AI models through an internal platform developed by the Swedish government.

This platform should help avoid potential data leaks and control what information is shared with the chatbot and the company.

Furthermore, variations of these models are included in Copilot, meaning that employees should still have access if their government account grants it.