On 26 October, the law firm Vinge published a review of supervision so far. Five of thirteen supervisory authorities had opened cases or made decisions, and eight had not. The Swedish Food Agency sent a request for information to 98 operators assessed as being required to notify. The County Administrative Board of Västra Götaland closed two cases against municipalities with remarks. No administrative fines had been decided, and no supervisory authority had carried out a security audit or a security scan. Vinge states that the information reflects the situation at the time of its review and that parts of the material are confidential.
Novudom Weekly – Week 44, 2026
On 26 October, the law firm Vinge described how supervision under the Swedish Cybersecurity Act has begun without any administrative fine having been decided. On the same day, the Government Offices of Sweden blocked ChatGPT and Claude on work computers. On 28 October, CERT-SE warned of a critical flaw in Brotli that CISA has added to its catalogue of exploited vulnerabilities. The common thread this week is that management needs to be able to show what is in use, who owns it and which suppliers can reach data. The letter ends with this week's bright spot.
Five of thirteen supervisory authorities have opened cases under the Swedish Cybersecurity Act and no fines have been decided
First establish whether your organisation falls under the Swedish Cybersecurity Act. For those affected, the next step is to ask a named owner to confirm that the notification has been made and that the documentation of security measures can be handed over at short notice. The cases so far mainly concern notification and information, so start there.
The Government Offices of Sweden block ChatGPT and Claude on work computers and steer staff to approved tools
According to Techtidningen on 26 October, the Government Offices of Sweden are blocking ChatGPT and Claude on work computers as of that day. Staff are directed to approved tools, including Microsoft Copilot. Director of digitalisation Mikael Wettercrantz is reported as saying that use without an agreement can have consequences for information security. Some employees reach the models through the Government Offices' own AI platform. The Swedish Tax Agency and the National Board of Institutional Care have similar blocks. The claim that the Swedish Enforcement Authority is considering a block is unconfirmed. The information comes from Techtidningen, which in turn cites the radio news programme Ekot.
Ask your head of IT which AI services employees use today, which ones are covered by an agreement, and who decides whether a service may be used. A block without an approved alternative risks pushing use onto private devices, so tie the decision to a tool employees can actually work in.
CERT-SE warns of critical flaw in Brotli that CISA has added to its catalogue of exploited vulnerabilities
On 28 October, CERT-SE published a warning about CVE-2026-111497 in the compression library Brotli, with a severity of 9.2 out of 10. An unauthenticated attacker can cause denial of service and possibly execute arbitrary code. CERT-SE states that CISA has added the flaw to its catalogue of known exploited vulnerabilities. BleepingComputer, which reported the flaw on 26 October, lists fixed versions of Brotli as 1.0.10, 1.1.1 and 1.2.1, and of Node.js as 22.23.4, 24.22.1 and 26.12.1. CERT-SE recommends that Brotli is not exposed to untrusted networks.
Ask your head of IT to confirm by Monday at the latest which systems and software you operate that use Brotli, including services delivered by others. Request an answer per supplier, since libraries are often embedded in products and the update then has to come from them.
Dustin's security chief says the breach began with access tokens exposed in a development mistake
Dustin's Group CISO Dennie Karlsson described how the breach unfolded in an interview in Techtidningen on 23 October. On 3 September, Dustin shut down parts of its IT after unauthorised access. According to Karlsson, the cause was a development mistake in which secrets and access tokens were exposed. Data was stolen and the company was subjected to extortion, but no encryption took place and the downtime was the result of Dustin's own shutdown. Old support tickets that remained in the ticketing system and were used as a knowledge base became a source of leakage. Among the lessons, Karlsson mentions that the team was too small during weeks two and three, and that companies should test what happens if five key people drop out.
Ask for a review of how secrets and access tokens are handled in development work and how old tickets and logs are purged. In your next exercise, rehearse a scenario in which five key people are absent for two weeks, and let the board see the outcome.
Nearly seven in ten people affected in breaches handled by Kroll were exposed through an external supplier
According to Kroll's Data Breach Outlook 2026, as reported by Help Net Security on 28 October, nearly seven in ten of the people affected in breaches handled by Kroll were exposed through an external supplier. Technology companies accounted for 6 percent of the cases but 60 percent of those affected. Around four in ten respondents cited supplier dependencies as a challenge. The basis is Kroll's own cases over the past year. The article reports neither the number of cases nor the method.
Ask your head of procurement or IT which three suppliers can reach the most personal data about your customers or residents, and who approved their access. Put supplier governance on the board's agenda once every six months.
68 percent of Swedish IT leaders trust their AI agents but 35 percent have full visibility of the supply chain
On 28 October, Dagens Infrastruktur reported on a Red Hat survey of 1,400 IT leaders in 14 countries, 100 of them in Sweden. Among the Swedish respondents, 68 percent trust that their AI agents operate securely and comply with regulations. At the same time, 35 percent have full visibility of their software supply chain, and 30 percent continuously follow up the risks at external AI suppliers. All Swedish respondents are reported to use AI agents in production. The article gives neither the name of the survey nor the field dates.
Ask the CIO to report to the board which AI agents are running in production, which suppliers they depend on, and when supplier risk was last assessed. Set as a target that every AI agent in production has an owner and a supplier assessment no more than twelve months old.
Proceeds from fraud in Sweden fell for the second year running to around SEK 5.7 billion in 2025
The Swedish Police estimate that the proceeds from completed frauds reported to the Police Authority were around SEK 5.7 billion in 2025, according to a news item on 20 April. That is a decrease of 10 percent from 2024, following a decrease of 16 percent between 2023 and 2024. Telephone fraud fell the most, from SEK 708 million in 2023 to SEK 320 million in 2025. The Police cannot determine which measures have done the most good, but point to police work against vishing, the cooperation with banks through Samlit, and information campaigns. The Police stress that the statistics should be interpreted with some caution and that attempted telephone fraud is increasing.
The trend is moving in the right direction, but the Police cannot single out which efforts had an effect, so keep going with all of them. Have your CFO and head of IT go through how you handle changed payment instructions. Fraud targeting company executives accounted for SEK 214 million in 2025.
Weekly reflection
Supervision has begun with questions about notification and documentation, the Government Offices have drawn a line on which AI services may be used, and at Dustin exposed access tokens led to systems being shut down. In all three cases, the question is whether management knows what is in use, who owns it and which suppliers can reach data. The NIS2 check takes three minutes and shows where you stand against the baseline requirements, including management responsibility and supplier governance.