Overview
- The Government Offices will block OpenAI’s ChatGPT and Anthropic’s Claude on staff computers, a change that takes effect Monday and names Microsoft Copilot the only approved AI assistant for routine government work.
- Mikael Wettercrantz, the Government Offices’ director of digitalisation, warned that using AI services without a formal contract can harm information security because employee queries are routed to external company servers.
- Copilot was allowed after an internal security review and raising of its privacy settings while ChatGPT and Claude did not receive comparable adjustments and were barred for general staff use.
- A limited number of employees will still be able to access the blocked models through an internal AI‑assistant platform the Government Offices built, though officials have not said how many users or what restrictions apply.
- The move follows similar bans at other Swedish agencies and reports of widespread 'shadow IT' use of about 10 AI services in one audit, and it leaves open broader questions about contracts, data flows and whether Copilot’s deployment includes Anthropic models.