legal obligation from Aug 3, 2026.

Hellooo!

Today I want to share something really important with you.

Since February 2025, AI literacy in European research institutes stopped being a nice-to-have. It became law. The supervision and enforcement rules apply from 3 August 2026 onwards.

What is article 4?

Article 4 of the EU AI Act requires every organisation that builds or uses AI systems to ensure their staff have a sufficient level of AI literacy. That’s not a recommendation or a best-practice guideline — it’s an enforceable obligation, and it applies to essentially every research institute on the continent.

Most researchers I speak to are vaguely aware of this. Far fewer know what to actually do about it.

The tricky part for research institutes? You’re sitting at a dual compliance point. If your team is developing an ML model, you’re an AI provider. If your grants office is using an LLM to draft funding reports, you’re an AI deployer. Article 4 obligations come at you from both directions.

Add to that the fact that a particle physicist using ML for data analysis has completely different literacy needs from an administrator using a document assistant — and one-size-fits-all training won’t cut it.

So what does compliance actually look like?

Three practical steps:

Map your AI use first. Before any training, you need to know which AI tools are in use, by whom, and in what context. Many institutes find this picture is broader than expected — tools adopted quietly at team level, long before any central strategy was in place.

Differentiate by role. A tiered approach works best: foundational awareness for all staff, plus role-specific competence training for those using AI directly in their work.

Document and iterate. The AI landscape changes faster than any training programme. What matters for compliance is an active, documented approach that gets reviewed and updated. The European Commission’s Living Repository of AI Literacy Practices is specifically designed to help with this.

The good news: compliance doesn’t require perfection. It requires intention, documentation, and progress.

I wrote up the full breakdown — including what the Apply AI Strategy adds to this picture and what institutions like CERN, Helmholtz, and EMBL are doing — in this post.

If you’re thinking through what AI integration looks like for your own work or institution, the European AI Compliance Hub is a good place to start.

Cheers,
Pranoti Kshirsagar Founder, The Science Talk

PS: If, apart from my email, you also like some proposal writing support, I think you’ll be happy about:

  1. Impact strategy for research proposalswhere I work with principal investigators who want professional science communication that matches the ambition and budget of their grant proposal. More details are on my website.
  2. Claude x GoogleSheets Connected - Claude’s native Google Drive connector does not allow access to GoogleSheets. So if you want to save time going back and forth then my (paid) guide👇 is perfect for you.

What’s inside:

✅ Complete OAuth 2.0 setup

✅ Copy-paste configuration

✅ 5 Claude prompts to use the moment you’re connected

✅ 7 common errors with exact fixes

✅ Security checklist to keep your credentials safe