Sweden Blocks ChatGPT and Claude at Its Government Offices

Sweden’s Government Offices, the Regeringskansliet, are blocking ChatGPT and Claude on staff work computers from Monday and want staff to use approved services, including Microsoft Copilot, Sveriges Radio’s news programme Ekot reported on 23 October 2026. Ekot also found blocks at the Swedish Tax Agency and at the National Board of Institutional Care, known as Sis.

Digitalisation director Mikael Wettercrantz declined an interview with Ekot but wrote on the staff intranet that using AI services without agreements can have “konsekvenser för vår informationssäkerhet”, which translates as consequences for our information security (our translation).

Business Pill · LEAST PRIVILEGE

A short Business Engineer explainer on least privilege: giving an AI agent only the access its task needs. It teaches the general idea only and says nothing about Sweden or any institution in this piece.

What Sweden’s Government Offices Did

• The block. From Monday, ChatGPT and Claude can no longer be reached from staff work computers, according to Ekot.
• The approved route. The block is there so that staff use the services the Government Offices have approved, among them Microsoft’s Copilot. Ekot says internal information shows security in Copilot was also raised after a review; the press service would not give details, citing security.
• An exception. The press service says some staff will keep access to an AI-assistant platform the Government Offices built themselves, through which they can reach ChatGPT and Claude.
• An earlier step. The Government Offices had already restricted use of DeepSeek, Ekot reports.

Ekot’s survey shows the Government Offices are not alone. Blocks are in place at the Tax Agency and at Sis, and the Enforcement Authority, Kronofogden, is preparing its position. At the Transport Administration, Ekot reports, internal audit found widespread and risky use of around ten AI services earlier this year, and wrote of “problematisk användning av öppna AI-tjänster”, problematic use of open AI services (our translation).

Who Is Restricting Which AI

We put Sweden’s step next to restrictions reported this year in the United States and at banks. The scopes differ: some cover staff using chatbots, one covers a vendor in government IT, and one covers outside agents inside customer accounts.

Sources: Sveriges Radio Ekot and Computer Sweden, 23 October 2026; Reuters and Fierce Healthcare, 2 March 2026; our pieces on the DHS DeepSeek order and on the Axios bank report. By our count, Ekot names three Swedish state bodies with blocks in place and one preparing a decision, and four of the five banks Axios names are blocking or restricting outside agents.

What the Cases Share

In each case where the reports name a replacement, staff are pointed to a tool the institution has chosen rather than to open consumer services: Copilot and an in-house platform in Sweden, ChatGPT Enterprise and Gemini at HHS, and StateChat at the State Department.

The stated reasons differ. The Government Offices’ digitalisation director cites agreements and information security; the US Claude phase-out followed a dispute over military use, Reuters reported; the DeepSeek order is a supply-chain exclusion; and the banks’ concern, per Axios, is agents logging in with customers’ passwords.

As we read it, the common thread is who controls the channel, not one shared reason. In Sweden the stated line runs between services with and without agreements; in the US Claude case, the State Department told Reuters it was taking immediate steps to implement the president’s direction to cancel Anthropic contracts. In both, staff are left with tools the institution has chosen.

Connecting the Coverage

Our bank agents piece reported Axios’s finding that Morgan Stanley has begun blocking third-party agents while JPMorgan, Bank of America and Citi are also restricting them. Our DeepSeek order piece covered the Homeland Security order telling civilian agencies to remove DeepSeek products from their IT systems. And our Pentagon piece reported a Defense Department statement that it “has ceased the use of Anthropic products.” What’s new: a European government office is blocking ChatGPT and Claude for staff from Monday and pointing them to Copilot.

The Structural Read

As we read it, the block does not end access to the models: the press service told Ekot some staff keep reaching ChatGPT and Claude through the Government Offices’ own AI-assistant platform, so the line is drawn around the channel, not the model.

As we read it, Copilot is the approved route in this case: Ekot reports staff are steered to approved services including Microsoft’s Copilot, whose security was raised after a review.

As we read it, the reasons in the table differ while the outcome is similar: information security in Sweden, a dispute over military use in the US Claude case per Reuters, a supply-chain order for DeepSeek, and customer credentials at the banks per Axios.

Three Implications

FOR AI VENDORS Ekot reports Swedish staff are steered to approved services including Copilot, and Fierce Healthcare reported HHS kept ChatGPT Enterprise and Google Gemini available; in both cases the product staff may use is the one the institution approved.

FOR IT AND SECURITY TEAMS Ekot reports Trafikverket’s internal audit found widespread and risky use of around ten AI services, and wrote of problematic use of open AI services (our translation).

FOR BANKS Axios reported that agents generally log in with the customer’s own credentials, and four of the five banks it names are blocking or restricting outside agents, by our count.

The Business Engineer Lens

This story maps onto the Business Engineer framework Inside Anthropic’s Permission Layer.

The analysis describes the governance layer this way: “It is where decisions are made about who gets access to frontier intelligence, under what conditions, and for which purposes.”

As we read it, the Swedish block is that layer inside one employer: the Government Offices decide which AI services staff may reach, and through which approved channel. The analysis was published in June 2026 about Anthropic; it does not discuss Sweden or workplace blocks.

What Is Not Established

Ekot’s report does not say how many staff are affected, when the Tax Agency and Sis introduced their blocks, or whether those blocks cover exactly the same services. The press service did not detail the Copilot security changes. The reports we read from March do not say whether every US department has completed the Claude phase-out. Dagens industri’s report is behind a paywall and we read only its opening, which relays Ekot; we did not contact the Government Offices, OpenAI, Anthropic or Microsoft. Nothing here is investment advice.

The Bottom Line

Sweden’s Government Offices are blocking ChatGPT and Claude on staff computers and steering staff to approved tools such as Copilot, Ekot reported, with similar blocks at two other Swedish agencies. Set next to the US Claude phase-out, the DeepSeek order and the banks’ limits on outside agents, the pattern as we read it is that the tool staff may use is the one the institution has approved.