AI News · Edition 77 ·
Which AI your staff may use is now a management decision
Sweden's government offices block ChatGPT and Claude but keep Copilot. Plus: a Node.js and Python security flaw, and an economist's 'zero payoff' verdict.
The one that matters
Sweden's government offices cut staff off from ChatGPT and Claude but keep Microsoft Copilot
What happened
Sweden's Government Offices, the organisation that serves the prime minister and the ministries, have blocked staff from using ChatGPT and Claude. Microsoft Copilot stays available. Several outlets have reported it. This is one large employer deciding which AI assistants its own people may use at work. It is not a ban for Swedish citizens or Swedish companies.
We won't guess at the exact reasoning in Stockholm. The pattern, though, is one that many organisations are reaching: choose one AI assistant, run it under a proper business arrangement, and close the side doors.
What it means for your business
Your staff are very likely using AI already, whether you approved it or not. In a 600-student school, it might be teachers drafting report-card comments in a free chatbot with student names and marks pasted in. In a 30-bed clinic, it might be the front desk tidying discharge summaries or insurance letters. In a workshop with 40 staff, it is usually the accounts or purchase person summarising supplier quotes and customer emails.
None of this is sinister, and much of it is useful. The problem is control. On free or personal accounts you have no admin view and no record of what was shared. Depending on the account's settings, the provider may also be able to use those conversations to improve its models. Suppose what was pasted includes patient details, student records, Aadhaar or SIN numbers, or customer bank details. Under India's Digital Personal Data Protection Act or Canada's privacy laws, the responsibility sits with your organisation, not with the employee who pasted it.
So the real lesson is that 'which AI tool?' has become a management decision, like choosing an accounting package. Business plans for the main assistants typically cost roughly US$20–30 per user per month, and you rarely need one for everyone. For the workshop, six office staff might come to roughly US$120–180 a month. For the school, start with the office team and the coordinators who write the most, not the whole staff room.
What it does NOT mean
- It does not mean ChatGPT or Claude are unsafe, or that Copilot is safer by nature. A government picks tools around its own contracts and security rules, and those are not yours.
- It does not mean you should move to Microsoft. If your office already runs on Microsoft 365, Copilot is the natural fit because it works inside the mail, files and permissions you already have. If you run on Google Workspace, Gemini is the equivalent. If you use neither, a business account with ChatGPT or Claude may suit you better.
- Watch who quotes this story at you. Expect some resellers to present it as proof that Copilot is the 'approved' choice. It was one employer's decision. Any headline that shrinks it to 'Sweden bans ChatGPT' overstates it too.
What a sensible owner should do this month
- Ask before you make rules. Send staff a short, no-blame question: which AI tools do you use, for what, and have you ever pasted in names, phone numbers, medical or financial details? A week of honest answers tells you more than any vendor demo.
- Pick one approved tool on a business account. Match it to the software you already pay for, and switch on the admin and data controls the business plan offers.
- Write a one-page rule. List what never goes into any AI tool: patient records, student marks with names, ID numbers, bank details, salaries, customer pricing. Then list what is fine: drafting notices, rewording emails, summarising your own non-confidential documents.
- Don't over-engineer the blocking. A government network can block websites centrally. A 40-person workshop usually can't, and doesn't need to. If the approved account is easier to use than the free one, staff will mostly switch on their own.
- Review in 90 days. See who actually uses the paid seats, and cancel the ones nobody touches.
If you do only one thing, do the first. Knowing what your people already do with AI is the cheapest risk check you will run this year.
Also worth knowing
How this briefing is put together
Every morning we read the day’s AI announcements and reporting from the companies themselves and from the technology press, then pick the handful that actually change something for a working business. The analysis is ours and it is written for owners and managers, not engineers. Every story links to its original source — read them, and disagree with us where we’ve got it wrong.