EU Buys Time for Schools While Models and Agents Accelerate

 



1. EU AI Act Gets Its First Major Overhaul

  • On May 7, EU institutions reached a provisional agreement on the Digital Omnibus on AI — the first set of amendments to the AI Act since it passed in June 2024.
  • The biggest change: compliance deadlines for high-risk AI systems (Annex III) pushed back 16 months, from August 2026 to December 2027, giving companies meaningful runway.
  • Two new prohibitions added, effective December 2026: AI-generated non-consensual intimate imagery and child sexual abuse material are now explicitly banned.
  • Penalties tightened — Article 25 breaches now carry fines of 3% of worldwide turnover; formal adoption expected this month with amendments entering force in July.

2. EU Launches Tech Sovereignty Package to Break US/China Dependency

  • The European Commission proposed the Cloud and AI Development Act (CADA), a sweeping framework to reduce the EU's reliance on US and Chinese cloud and AI providers.
  • The package establishes tiered "sovereignty levels" for cloud services, requiring more EU-controlled infrastructure for sensitive workloads.
  • Separately, the EU's €20 billion plan for five AI mega-data centers is reportedly stalling on funding and partner alignment — raising questions about how CADA ambitions will be funded.
  • Context: Europe is increasingly citing geopolitical risk ("nobody has a kill switch") as the driver, not just economic competition.

3. OECD Digital Education Outlook 2026: GenAI Helps Only When Taught Well

  • The OECD released its Digital Education Outlook 2026, the most authoritative international benchmark for AI in classrooms.
  • Key finding: GenAI supports learning when teachers provide clear pedagogical scaffolding — but without guidance, students outsource thinking entirely and gain nothing measurable.
  • US states are responding fast: 134 bills in 31 states are now tracking AI-in-education legislation, up sharply from 2025, covering privacy, parental consent, and human oversight requirements.
  • New York City is rolling out one of the most consequential AI governance frameworks in US K-12 — covering 1,700+ schools — with a compliance playbook released this month.

4. Microsoft Launches 7 MAI Models at Build 2026

  • At its Build developer conference, Microsoft unveiled a family of seven in-house models under the MAI brand, covering reasoning (MAI-Thinking-1), coding (MAI-Code-1-Flash), image generation (MAI-Image-2.5), transcription (MAI-Transcribe-1.5), and voice (MAI-Voice-2 in 15 languages).
  • MAI-Thinking-1 reportedly outperforms Claude Sonnet 4.6 in blind evaluations on software engineering benchmarks; MAI-Code-1-Flash ships integrated into GitHub Copilot and VS Code.
  • Microsoft explicitly distances itself from distillation: "We don't distill from other labs" — and introduces Frontier Tuning for enterprise customization, claiming up to 10× efficiency on proprietary workflows.
  • The release signals Microsoft's intent to reduce dependency on OpenAI and compete directly on model quality and price.

5. Google and Meta Race to Ship Personal AI Agents

  • Google is internally testing Remy, a personal AI agent embedded in the Gemini app, designed as an autonomous "24/7" assistant for work, school, and daily life — it monitors activity and learns user preferences over time. Google canceled Project Mariner to focus resources here.
  • Meta is building Hatch, a competitive personal agent currently running on Anthropic's Claude models, with an internal test by end of June; it will switch to Meta's own Muse Spark model at launch.
  • Meta is also building an Instagram shopping agent that lets users buy items from Reels without leaving the app — a direct TikTok Shop play timed before Q4.
  • Both are chasing Anthropic and OpenAI, which already have agent products in user hands — a gap the article describes as "clear and growing."

Why it matters for educators: The OECD's warning that unsupported GenAI use produces zero learning gains — combined with the EU deadline extensions and US state legislative surge — means schools have a narrow but real window to build AI policies before compliance mandates catch up to them.

Sources

Comments

Post a Comment