Skip to main content Scroll Top

November & December 2025

NOVEMBER DECEMBER 2025

AI Governance and Tech Review November & December 2025

By Jana Schรถneborn
๐Ÿ‡ช๐Ÿ‡บ ๐—ง๐—ต๐—ฒ ๐—˜๐—จ ๐—”๐—œ ๐—”๐—ฐ๐˜ ๐—ฎ๐—ป๐—ฑ ๐—ฟ๐—ฒ๐—น๐—ฎ๐˜๐—ฒ๐—ฑ ๐—˜๐˜‚๐—ฟ๐—ผ๐—ฝ๐—ฒ๐—ฎ๐—ป ๐—ฟ๐—ฒ๐—ด๐˜‚๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป:
  • Within this timespan, the AI Office did not onlyย start the working groupsย on the Code of Practice on Transparency obligations, but already produced aย first draftย containing almost hands-on guidance on how to label and mark AI-generated content.
  • Most talked about and most relevant for 2026: The Digital Omnibus on AI Regulation proposal has first beenย leakedย and thenย publishedย in its full extend: Meaning to simplify and unify regulation, tying obligations for high-risk AI systems closer to the provision of standards, it still raised quite a few questions. Here I want to mention two particularly interesting ones: 1. the proposedย change to GDPRย in order to facilitate AI training and 2. the shifted (or possibly not?) timelines for high-risk requirements, withย backstop compliance datesย in December 2027 (for systems listed in Annex III) and August 2028 (Annex I).
  • The Danish NGO Institute for Human Rights published aย Guide and a template for the Fundamental Rights Impact Assessmentย – one possible and surely helpful and quite extensive interpretation of what the AI Act (article 27) requires.
  • Lastly, the AI Office launched itsย Whistle Blower app.
๐Ÿ“š ๐—”๐—œ ๐—น๐—ถ๐˜๐—ฒ๐—ฟ๐—ฎ๐—ฐ๐˜†:
  • One of the clear points of the Omnibus refers to a topic I very much care about, AI Literacy, demoting it from an obligation for organisations to a recommendation, while shifting the responsibility to governments. It was such an achievement to position this obligation differently than other regulations – I will write about the implications of this soon!
  • The Rolling Stones Magazine (yes, just that one) published a well-written (and -researched) article onย AI flop in academia.
๐Ÿ‡บ๐Ÿ‡ธ ๐—จ๐—ฆ ๐—ฅ๐—ฒ๐—ด๐˜‚๐—น๐—ฎ๐˜๐—ถ๐—ผ๐—ป & ๐—š๐—ผ๐˜ƒ๐—ฒ๐—ฟ๐—ป๐—ฎ๐—ป๐—ฐ๐—ฒ:
  • THE one over-arching US-topic was clearly theย Executive Oderย signed by the president aiming (again) to end state-regulation regarding (almost) anything AI.
AI Governance globally:
  • India released its โ€œIndia AI Governance Guidelinesโ€ defining the countryโ€™s own position on AI Governance, including a focus on trust, the commitment to establish interdisciplinary committees within the government (including the AI Governance Group and the AI Safety Institute) alongside practical guidelines.
  • Canada launched their AI register forย AI use cases used by the government.
  • Sirak Abrahaย brought my attention to theseย newsfrom the UK with relevance for all banks using Agentic AI.
๐Ÿ“„ ๐—ฃ๐—ฎ๐—ฝ๐—ฒ๐—ฟ๐˜€ & ๐˜€๐˜๐˜‚๐—ฑ๐—ถ๐—ฒ๐˜€:
  • Capgemini & the World Economic Forum published this brightย reportย on Governing Agents.
Tech-developments:
  • Not too surprisingly after a few decisive staffing choices and LeCunโ€™s sustained criticism of the very basis (current LLMs) of the race Meta wants to compete in, he LeCunย announced to be leaving the companyย and instead work on his own startup.
  • Antrophic published an insightful report on a campaign detected in specific but allowing conclusions for the state of cyber security in general includingย jailbreaks, human oversight (works both ways), infiltration and much more.
  • Code Red at OpenAIย following a seemingly too-good Gemini-release.
  • โ€ฆno good review without scary Altman-content (he might get his own category soon): While I also turn to the internet (there was no ChatGPT when my children were babies) for some parenting choices, I do hope Altman is not representative in stating (without shame!) that heย could not imagine caring for a baby without ChatGPT.