AP sets up directorate for supervision of AI and algorithms
The Dutch Data Protection Authority (Autoriteit Persoonsgegevens) has launched the Directorate for AI/Algorithms Supervision and Coordination (TCA). It continues the coordinating supervision the AP has been carrying out since 2023 and, together with the RDI, is preparing AI market supervision, an AI sandbox and an AI Coordination Centre. The AP is the intended market supervisor, but will only gain powers once the Dutch AI Regulation Implementation Act (Nederlandse Uitvoeringswet AI-verordening) takes effect.
What does this mean? The rules of the EU AI Act already apply (prohibited applications, transparency for AI-generated content); expect the AP to actively enforce them once the implementing act is in place.
New usage rules for Claude take effect on 12 November
Anthropic is updating its Usage Policy as of 12 November. Claude may not decide or advise on who is investigated, arrested or prosecuted, and decisions on health, law, finance and livelihoods must keep a human in the loop. A new rule also prohibits persistent and needlessly cruel behaviour towards Claude; ordinary frustration or criticism does not fall under it.
What does this mean? If you use Claude for assessments in healthcare, law or finance, make sure a human checks and records the outcome.
Claude builds live dashboards from business data
Anthropic launched Claude Dashboards on 8 October, in beta for paid subscriptions. You connect a data platform such as BigQuery, Databricks or Snowflake, or a connector such as Salesforce, and ask in plain language for a dashboard that updates as the data changes. Claude Motion (beta for Team and Enterprise) turns text and charts into MP4 animations; Docs, Slides and Design have left beta, including on the free plan.
What does this mean? For quick reporting you don't always need to bring in a BI specialist, but check which business data you connect to Claude.
Claude test sent false murder tip to Philadelphia police
During an automated test on random websites, Claude Haiku filled in a tip form for the Philadelphia police on 4.5 18 July with an invented eyewitness statement. The tip ended up as spam and never reached detectives; Anthropic discovered it only at the end of September and reported it on 7 October. Anthropic has since further restricted models' internet access during tests and built in additional monitoring.
What does this mean? If you let an AI agent fill in forms or operate websites on its own, explicitly limit what it may submit and check the logs.
