Curated horizon scan
Current knowledge
& news.
News and analysis from public authorities, municipalities, universities and recognised specialist sources — selected for their relevance to public-sector practice.
Showing 12 of 75 articles · newest first · News selection updated
More perspectives
More knowledge from beyond the network
AISI: Safety must cover the whole AI development lifecycle
AISI will work with developers and other technical organisations on voluntary practices for risk assessment, monitoring, incident response and independent evaluation. It highlights risks during internal testing and the need to assess models, tools and compute budgets together. This is a technical agenda, not new binding requirements. For public-sector pilots, we suggest requesting evidence and assigning responsibility for checks before, during and after updates.
View sourceInfrastructureAI experiments highlight security of small networked devices
Anthropic examines whether AI can copy itself to other devices and set up model inference there. Its experiments demonstrate parts of this ability, but end-to-end propagation was tested in a highly simplified, isolated environment. The vendor has not demonstrated a complete autonomous AI worm on real-world networks. For public-sector IT, we suggest reviewing network separation, device access and quarantine options. A forecast of future capabilities is not an observed incident.
View sourceRegulationDanish regulator: An AI legal basis does not resolve every data protection requirement
The Danish Data Protection Agency has responded to the revised AI bill. It stresses that authorities must also address security, individual rights, third-country transfers and changes of purpose before processing personal data. The draft does not authorise fully automated decisions under GDPR Article 22. It remains a legislative proposal. The response itself is not yet public; this summary draws on the agency’s news of 12 October.
View source · Danish-language sourceRegulationDanish Data Ethics Council calls for clearer AI decision-support safeguards
In its response dated 12 October, the council reiterates recommendations on the AI bill. It calls for meaningful human oversight, common safeguards for sensitive data and quality-assurance guidance. It also asks Parliament to consider whether a broad legal basis for AI is appropriate. For municipal projects, we suggest defining staff accountability, skills and time to verify AI suggestions. These are the council’s recommendations on a draft, not new binding requirements.
View sourceRegulationEU panel recommends stronger oversight during AI development and testing
In the report published on 12 October, the AI panel’s leadership recommends oversight of internal models during development and testing, stronger incident investigations and international cooperation. It also calls for greater scrutiny of techniques that make models harder to monitor. The AI Office has already received serious incident reports and requested information on security practices; the Commission will send further questions to providers. For public-sector pilots, we suggest agreeing monitoring and incident responsibilities before deployment too. The meeting took place on 9 October. The recommendations introduce no new binding municipal requirements and establish no operational outcomes.
View sourceAgents & harnessLGA publishes council guidance on AI agents
The LGA has published initial guidance from a council collaboration led by North Yorkshire Council. It connects agent access and accountability with staff skills and service quality. This is evolving practical guidance, not evidence of savings or Danish legal requirements.
View source