Sweden has barred government officials from using OpenAI’s ChatGPT and Anthropic’s Claude for official work, citing risks to information security, and has directed staff to government-approved tools such as Microsoft Copilot instead.
The block takes effect immediately, Hiru News reported at 8.35 p.m. Colombo time on Monday, attributing the account to foreign media.
Sweden’s Chief Information Officer, Mikael Wettercrantz, said public servants should from now on use only AI tools the government has approved, and that using unapproved tools could pose a threat to the country’s information security.
What prompted it
A recent internal audit at the Swedish Transport Administration found a number of problems with the way officials were using AI — entering sensitive state information into tools that are not secure, and then using what those tools returned.
The restriction is not absolute. A limited group of officials can still reach ChatGPT and Claude through a dedicated AI platform the government has built itself for specific tasks.
International reporting on the decision, which this desk reviewed but does not cite as a source, adds that the block took effect on 26 October, that the Government Offices confirmed it in an email to the Swedish public broadcaster SVT, and that Microsoft Copilot was permitted because the Government Offices reviewed it and tightened its security settings — a review not carried out for the other two services. Those reports also say Sweden’s Tax Agency and its National Board of Institutional Care had already imposed similar restrictions.
Why it matters here
The decision is a working example of the governance question Sri Lanka’s own institutions are being pressed on. At the Central Bank’s oration this month, the IMF told Sri Lanka it is closer to an AI future than it thinks, with the discussion turning on trust and on the conditions under which public bodies should rely on these systems.
Sweden’s answer is not a rejection of the technology — it keeps one commercial assistant, builds its own platform for the rest, and draws the line at where official data goes. That is a procurement and data-handling judgement rather than a verdict on the tools.
A note on sourcing and names
Hiru News is the only verified newsroom in this desk’s source list to have filed on this, and it attributes the story to unnamed foreign media rather than to a Swedish government release. No Swedish government statement was obtained directly for this report.
Hiru renders the Chief Information Officer’s name as “Michael Wetterkrants”. International reporting gives it as Mikael Wettercrantz, and describes him as CIO and Director of Digitisation; this report uses that form.
What is not established
No filing reviewed gives the legal or administrative instrument behind the block, says whether it is permanent or under review, or states how many officials are affected. Neither the number of staff retaining access through the government platform nor the identity of that platform’s supplier is reported. There is no reported comment from OpenAI or Anthropic, and no filing says whether either company was consulted or notified before the decision. The findings of the Transport Administration audit have not been published in the reporting reviewed.
