A block with a door left open
Sweden’s Government Offices — the Regeringskansliet, the civil service body that supports the prime minister and every ministry — will block ChatGPT and Claude on staff work computers from Monday, Sveriges Radio’s Ekot reported on 23 October. Staff who need AI are directed to services the Government Offices have approved, Microsoft Copilot among them.
The block is not absolute. Some staff keep access to an AI platform the Government Offices built themselves, and ChatGPT and Claude can be used through it. The distinction is contractual rather than technical: what is being cut off is direct use of the vendors’ own services, not the underlying models.
Use of DeepSeek had already been restricted before this decision.
The reason given
Mikael Wettercrantz, the Government Offices’ director of digitalisation, declined an interview. In a comment on the internal intranet he wrote that using AI services without agreements in place can have consequences for information security.
Ekot also reports that the security of Copilot was raised following a review. The press office declined to give details, citing security grounds. That leaves the central question unanswered in public: whether the problem was something found in ChatGPT and Claude specifically, or the absence of a procurement agreement covering them.
Not the first Swedish agency
The Swedish Tax Agency and Statens institutionsstyrelse, which runs compulsory institutional care, have both introduced blocks of their own. The Swedish Enforcement Authority is preparing the question ahead of a decision.
The pattern has a precedent inside the Swedish state. An internal audit at the Swedish Transport Administration earlier this year found widespread and risky use of around ten AI services. “We have seen several examples of problematic use of open AI services,” the audit found.
Why it matters beyond Sweden
Government bodies across Europe are working out the same thing at once: that the question is not whether staff may use AI, but whose terms they use it on. A block paired with an in-house platform that reaches the same models is a procurement answer, not a capability one. It keeps the data inside an agreement the employer has signed.
That is also the model most likely to spread. It lets an organisation say it has restricted consumer AI services while continuing to use the models behind them, and it puts the vendor relationship where a public body can audit it.
What to watch
The Enforcement Authority’s decision is the next one due, and it will show whether this becomes a sector-wide position rather than a run of separate calls. The more revealing detail would be what the Copilot review actually found. The Government Offices have raised that product’s security and declined to say what needed raising, which is the one piece of information that would let anyone outside judge whether the approved tool is safer than the blocked ones or simply better papered.
