According to IRIB News, citing the Ekot news section; the Swedish government, citing information security and data protection considerations, has restricted the access of government agency employees to the artificial intelligence tools ChatGPT and Claude, and in return has allowed the use of Microsoft Copilot with the application of special security settings; a decision that once again raises the debate about data security, digital sovereignty and governments' dependence on big technology companies.
Swedish government offices, from Monday, October 26, are restricting access to the artificial intelligence services ChatGPT, owned by OpenAI, and Claude, owned by Anthropic, on employees' organizational computers. In contrast, the Microsoft Copilot tool, after security reviews and strengthening of protective settings, still remains on the list of allowed tools.
Mikael Vetrankrants, the digitalization manager of Swedish government offices, in explaining this decision, has stated that using artificial intelligence services without formal and specific contracts can put organizational information security at risk. According to the published report, the concern of Swedish officials is especially about the sensitivity of the information that employees place at the disposal of these tools, as well as how the data and the responses produced by artificial intelligence models are processed.
Data security; the main axis of the restrictions
The Swedish government's decision was adopted in conditions where the use of generative artificial intelligence tools in government institutions is expanding rapidly, but the legal and security frameworks proportionate to this trend have not yet fully taken shape in many organizations. When employees place administrative information, internal documents or sensitive data at the disposal of a foreign artificial intelligence service to receive an answer, how this data is stored, processed and possibly used becomes one of the main concerns of government agencies.
In such conditions, formal contracts with service providers can determine the limits of responsibilities, the obligations related to information protection and the conditions of data use. It seems the Swedish government, in its recent decision, has emphasized, in addition to the nature of the technology, the degree of organizational control over how it is used.
According to this report, some employees can still access ChatGPT and Claude through an internal artificial intelligence platform; however, government officials have not announced the details about the number of users and the conditions of this access.
The AI restriction has also reached other Swedish institutions
This approach is not limited to government offices. The Swedish Tax Agency and the National Board of Institutional Care of this country have also taken similar measures in the field of using artificial intelligence tools. The Swedish Enforcement and Debt Collection Agency is also examining a similar decision. On the other hand, an internal audit in the Swedish Transport Agency has reported the widespread use of about 10 artificial intelligence services that were accompanied by security risks.
These cases show that the rapid expansion of the use of artificial intelligence tools in administrative environments has not necessarily been accompanied by the drafting of proportionate guidelines and supervisory mechanisms.
The message of this decision for AI policymaking
Sweden's action is part of a broader trend of governments and public institutions reviewing their artificial intelligence use policies. The European Parliament had also previously disabled some artificial intelligence capabilities on its work devices; a move in which concerns about data protection played a role.
With the expansion of the use of artificial intelligence in executive agencies, universities, state companies and institutions providing public services, drafting clear rules for entering organizational information into foreign services becomes a serious necessity.
Sweden's experience shows that an effective policy in this field does not necessarily mean an all-encompassing ban on artificial intelligence; rather, it can rest on security assessment of services, setting specific contracts, determining users' level of access, protection of sensitive data and monitoring how the models are used.
The key point for policymakers is that the main issue is not merely choosing between ChatGPT, Claude or Copilot; rather, it is creating an enforceable framework for protecting public data, assessing the risk of artificial intelligence tools and preserving organizational control over information.
Source: IRIB News
