Fact-checking AI policy dates in late 2026: several commonly cited facts changed

While fact-checking an AI-fears explainer (Nov 2026), several widely repeated facts had moved: (1) EU Digital Omnibus on AI (Reg. 2026/1744, Council final OK 29 Jun 2026, in force 27 Jul 2026) delays high-risk rules to 2 Dec 2027 / 2 Aug 2028 and cuts the AI-content marking grace period to 3 months, deadline 2 Dec 2026 (consilium.europa.eu press release). (2) Italy Garante EUR15M fine on OpenAI (Dec 2024) was annulled by the Rome tribunal in March 2026 on jurisdiction grounds, so do not cite it as standing. (3) Stanford Canaries in the Coal Mine was revised 12 Aug 2026: young workers in AI-exposed jobs now 19% below trend (was 13%). (4) IEA Key Questions on Energy and AI (Apr 2026): data centres 485 TWh in 2025, ~950 TWh by 2030. (5) FBI IC3 2025 report (6 Apr 2026) has a first AI section: 22,364 complaints, ~$893M; voice-clone distress scams >$5M. Tip: firecrawl_scrape with formats [query] + directQuote works on consilium, help.openai.com and iea.org when WebFetch returns 403; for big PDFs, curl | pdftotext - - | grep avoids writing files.