Two governments have moved against specific AI tools on official computers in the same week, and the reasons are different enough to be worth separating.

On October 26, Sweden’s Government Offices began blocking ChatGPT and Claude on staff work computers. In the United States, the Department of Homeland Security has ordered civilian federal agencies to remove DeepSeek and to stop buying it. Neither move bans AI at work. Both are about which AI services an organization has a contract, and a trust relationship, with.

Sweden: no contract, no access

Sveriges Radio’s Ekot reported the Swedish decision, citing internal information, and trade coverage describes it as follows:

  • The block applies to ChatGPT and Claude on Government Offices (Regeringskansliet) work computers from October 26.
  • Staff are pointed to Microsoft Copilot, which remains approved. The office reportedly tightened Copilot’s security settings after an internal review. It has not said whether Claude models inside Copilot are enabled.
  • The digitalisation director, Mikael Wettercrantz, warned on the staff intranet that using AI services without a contract could affect information security. Reporting says the underlying problem is the absence of data agreements with OpenAI and Anthropic.
  • It is not a total shutdown. Some employees keep access to the two assistants through an in-house AI platform the office built, though the office has not said who or how many.
  • The Swedish Tax Agency and the National Board of Institutional Care are reported to have similar restrictions, and the Enforcement Authority is preparing one.

The key detail is that the objection is procedural, not a finding that either tool is unsafe. Without a contract that says where data goes and how it is handled, a government employee pasting a document into a public chatbot is a risk the office cannot manage. A tool that comes with an enterprise agreement can clear the same bar.

United States: a security order against DeepSeek

The American action is aimed at a Chinese AI provider. According to The Hacker News, ExecutiveGov and other coverage, a DHS directive requires civilian executive-branch agencies to remove DeepSeek products, including third-party products and services built using DeepSeek implementations, and bars new purchases through any buying channel. The stated concern is data security.

Some details are still unsettled:

  • When it was signed. Reports say the directive was signed this summer and only made public in October through a Federal Register notice. Outlets give different signing dates, June 26 in one account and July 24 in others, so we are not stating one.
  • Exceptions. The directive reportedly let agencies seek relief within 60 days. One outlet argues that window closed before the order became public, which would leave agencies no practical route to an exception. We could not confirm that from a primary source.
  • Scope. It covers civilian agencies but not national security systems, and reporting indicates the Department of Defense is not covered.

The “even in other apps” point matters most for contractors and vendors: the order reaches products that embed DeepSeek models, not only the DeepSeek app.

What this means for working professionals

Neither government is saying “don’t use AI.” Both are saying “know which AI, under which terms.” That is a useful checklist for any organization:

  • Lawyers and compliance teams. Expect procurement questions about data processing agreements, retention and training use, in addition to accuracy. A tool without a signed agreement may be blocked even if it works well.
  • Teachers and public-sector staff. Check your employer’s approved-tools list before using a consumer account such as ChatGPT or Claude with work material. Consumer and enterprise versions often come with very different terms.
  • Anyone building on open or foreign models. The US order shows that a model’s origin can become a procurement issue. If a product you sell to government customers includes a restricted model, you may have to remove it.

What to watch

  • Whether OpenAI or Anthropic signs agreements that satisfy Sweden’s requirements, and whether the block is lifted.
  • Whether other Swedish agencies, or other governments, adopt similar approved-tool lists.
  • Whether more countries follow the US lead on DeepSeek, and how contractors are expected to certify compliance.

The common thread is that for public-sector work, the contract is now part of the product.