Sweden’s Authorities Places of work are blocking direct browser entry to ChatGPT and Claude on workers computer systems whereas conserving Microsoft Copilot out there for official work.
The Swedish Authorities Places of work, which help the prime minister and authorities ministries, blocked browser entry to OpenAI’s ChatGPT and Anthropic’s Claude on workers computer systems on Monday, Oct. 26, in response to Sveriges Radio and an electronic mail the federal government despatched to SVT.
Officers need workers to make use of AI providers authorised by the Authorities Places of work. Mikael Wettercrantz, the Authorities Places of work’ director of digitalization, warned on the interior workers community that utilizing AI providers with out formal agreements may have “penalties for our data safety,” Sveriges Radio reported.
Individually, an inside audit on the Swedish Transport Administration recognized widespread, dangerous use of about 10 AI providers. The audit discovered issues involving each the sensitivity of data workers entered into AI instruments and the way they dealt with the ensuing responses, in response to Sveriges Radio.
Why Copilot Stays Obtainable
Microsoft Copilot stays authorised after a authorities assessment, with further safety measures launched following the evaluation. Officers haven’t disclosed the particular modifications.
The excellence facilities on organizational approval, contractual protections and safety controls, somewhat than establishing that one AI mannequin is inherently safer than one other. Formal agreements can outline how suppliers course of data and what protections apply when workers use AI instruments.
Some Authorities Places of work workers retain entry to ChatGPT and Claude by means of Intric, an authorised AI platform, SVT reported. The platform’s provider introduced in July that it had received a contract to offer its AI platform to the Authorities Places of work.
The experiences don’t specify what number of workers retain entry by means of Intric or element the safety controls in that deployment.
Extra Swedish businesses are tightening AI entry
The Authorities Places of work should not appearing alone. Sveriges Radio reported that the Swedish Tax Company and the Nationwide Board of Institutional Care have additionally restricted AI providers. The Swedish Enforcement Authority is contemplating its personal measures.
The Authorities Places of work had beforehand restricted use of DeepSeek, in response to Sveriges Radio. Elsewhere, the European Parliament disabled built-in AI options on staff-issued gadgets in February 2026 over cybersecurity and data-protection issues, TechRadar reported.
Extra must-read AI protection
What IT groups ought to assessment earlier than approving AI entry
The restriction applies to Authorities Places of work work computer systems, somewhat than customers or non-public companies throughout Sweden. For IT groups adopting generative AI, it provides a motive to assessment which providers workers use, which accounts they register with, and what data they submit.
Approval ought to cowl the particular service and deployment. Groups ought to doc which accounts, interfaces, integrations, and duties are permitted so workers perceive when a device is authorised for work. Entry by means of a private account shouldn’t be assumed to hold the identical protections as a company’s contracted service.
Earlier than permitting confidential data into an AI device, organizations ought to assessment contractual phrases, information retention, model-training provisions, administrative controls, and any related purposes. Workers want clear examples of what they could submit and which data requires further authorization.
An authorised different additionally must help the work workers truly do. IT groups ought to present an easy course of for requesting new capabilities, reporting unintended disclosures, and checking generated materials earlier than it informs consequential selections.
The sensible lesson is to make authorised AI use each comprehensible and workable. Blocking an internet site addresses one entry route; a helpful office coverage additionally establishes who can use every service, what information it could deal with, and who’s accountable for reviewing its output.