Compliance

Sweden blocks public AI chatbots: lessons for ASEAN

nPro Team · 28 October 2026 · 3 min read

Sweden's Government Offices have blocked ChatGPT and Claude on staff computers. What the move means for AI governance, shadow AI and PDPA compliance in Malaysia.

On Monday 26 October, the Swedish Government Offices, which support the prime minister and cabinet, blocked ChatGPT and Claude on staff work computers. Microsoft Copilot, the only AI tool to pass the department's security review, remains available. No breach has been confirmed. This is a governance decision, and one that CIOs and DPOs in Malaysia should study closely.

What happened

According to Decrypt, citing Sveriges Radio, digitalisation director Mikael Wettercrantz told staff that using AI services without a formal contract can have "consequences for our information security". He said there had been "several examples of problematic use of open AI services", covering both the sensitivity of information entered and how AI responses were handled. Some staff can still reach the blocked tools through an internal AI platform built by the department. Sweden's Tax Agency and the National Board of Institutional Care have introduced similar restrictions.

The same week brought a reminder of what is at stake. On 22 October, Cybernews reported that 302.ai, an AI aggregator operated by Hong Kong-based Univerads Technology Limited, had left an Azure storage bucket publicly accessible. It held about 17.6 million files (22.72TB), including more than 144,000 chat logs, personal details and private documents users had uploaded for analysis. Researchers found it on 2 August, and it was observed closed on 8 September.

Why it matters for Malaysia and ASEAN

Every prompt sent to an external AI service is data leaving your control. Once it sits on someone else's infrastructure, its protection depends on their configuration, as the 302.ai case shows.

In Malaysia, the stakes are higher than they were two years ago. The Personal Data Protection (Amendment) Act 2024 is now fully in force. The maximum fine for breaching the data protection principles has risen from RM300,000 to RM1,000,000, controllers must notify the Personal Data Protection Commissioner of personal data breaches, data processors are directly bound by the Security Principle, and organisations above certain thresholds must appoint a DPO. If an employee pastes customer records into an unapproved chatbot, the organisation still answers for that data.

Five practical takeaways

  1. Decide which AI tools are approved, and on what terms. Sweden's test was simple: no formal agreement on data protection and processing, no access.
  2. Offer a sanctioned alternative. Sweden kept Copilot and an internal platform. A ban with no alternative pushes staff towards personal devices.
  3. Find your sensitive data first. You cannot protect what you have not located, so map where personal and confidential data sits on endpoints.
  4. Control the exit routes. Copy and paste, browser uploads, USB drives and screenshots are common ways sensitive data reaches unapproved services.
  5. Keep evidence. With breach notification now mandatory under the PDPA, you need records of what left, when, and through which channel.

How a self-hosted platform helps

nPro is a self-hosted security platform that runs on-premises, in a private cloud or in a public cloud, so your security data stays on infrastructure you choose. nPro DLP can block sensitive content at the main exit points: clipboard protection, cloud upload prevention through a browser extension, USB and device control, and screen capture control. Sensitive data discovery and print monitoring show where sensitive data lives and when it is printed. Every DLP event lands in the same database and dashboard as nPro XDR, SIEM and EMS, so your team investigates one timeline rather than several. Learn more about nPro DLP.

Start with what you can see

Before you write an AI policy, find out what your staff are already sharing. nPro's free 30-day Data Exposure Audit shows where sensitive data sits and how it moves across your endpoints. Request your Data Exposure Audit.

Frequently asked questions

Why did Sweden's Government Offices block ChatGPT and Claude?

The Swedish Government Offices blocked ChatGPT and Claude on staff computers from 26 October 2026 for information security reasons. Officials said using AI services without a formal contract on data protection and processing creates risk, and cited several examples of problematic use. Microsoft Copilot, which passed a security review, remains approved.

What is shadow AI?

Shadow AI is the use of AI tools that an organisation has not approved or assessed, such as staff pasting work documents into a public chatbot. The risk is that sensitive or personal data leaves the organisation and is stored on infrastructure it does not control.

How can organisations in Malaysia reduce the risk of data leaking into AI tools?

Organisations can approve specific AI tools under formal agreements, provide a sanctioned alternative, locate sensitive data on endpoints, and use data loss prevention (DLP) controls on clipboard, browser uploads, USB devices and screen capture. Keeping logs of these events supports breach notification duties under Malaysia's amended PDPA.

Sources

Related reading

See how nPro implements this

The documentation covers configuration, agent deployment and the current status of each capability.

View documentation More articles