EU AI Initiatives

Ian Hogarth's opinion piece in the Financial Time echos previous remarks by Emmanuel Macron and Mario Draghi: technology drives economic growth and Europe is missing some ingredients that would enable the emergence of large technology companies. Europe has the required talents: lots of AI breakthroughs were produced in Europe by Europeans, but with .... US funds (e.g. at Google-DeepMind in London or Meta-FAIR in Paris). Ian points to a lower tolerance for risk in Europe than in the US, both from entrepreneurs and (perhaps more importantly) from investors. There is that. But there is another important factor: almost all of the fundamental innovations in AI of the last dozen years did *not* come from startups. They came from well-funded industry research labs belonging to large and highly-profitable companies: Google, Meta, Microsoft, and a few others. DeepMind would *never* have survived, let alone deliver breakthroughs, without being bought by a large company like Google. Their original business model as an independent company was never going to fly, in part because long-term research is expensive, and in part because they were overly optimistic about their time-line to AGI (their original plan for AGI based on RL was a complete failure). Why haven't large European group started ambitious AI research labs in the vain of Google Brain, DeepMind, FAIR, or MSR? European comapanies used to have world-class research labs, but not anymore. In the heydays of Bell Labs, IBM Research, Xerox PARC, and Microsoft Research, there were such labs Europe in the 1980s (e.g. Phillips Labs, Siemens, France Telecom, Alcatel). But they never valued research scientist careers like American tech companies have re-learned to do it, starting with MSR in the laste 90s. European industry research labs became shadows of their former selves. But interestingly, they caused some interesting spin-offs to exist: the most valuable European tech company is ASML, which was created on the remnant of Phillips Labs. The existence of ambitious industry research labs has an incredibly positive effect on the R&D and startup ecosystem. I witnessed this effect first-hand with the creation of FAIR-Paris in 2015: it almost single-handedly jump-started the AI startup ecosystem in Paris (which is now the most vibrant in Europe today). The existence of FAIR-Paris, and later the Parisian branch of DeepMind sent a message to the young aspiring scientists: you can have a career in AI research in Europe, and outside of academia. It motivated and lot of talented students to pursue graduate studies, to learn how to do research by doing a PhD. FAIR-Paris contributed to this by hosting PhD students in residence. FAIR graduates a dozen PhDs in an average year. They have gone on to do wonderful things in the European ecosystem. Some have founded AI startups and raised large amounts of capital .... but often from US investors. https://lnkd.in/eh4FtSsS

The real challenge in AI today isn’t just building an agent—it’s scaling it reliably in production. An AI agent that works in a demo often breaks when handling large, real-world workloads. Why? Because scaling requires a layered architecture with multiple interdependent components. Here’s a breakdown of the 8 essential building blocks for scalable AI agents: 𝟭. 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀 Frameworks like LangGraph (scalable task graphs), CrewAI (role-based agents), and Autogen (multi-agent workflows) provide the backbone for orchestrating complex tasks. ADK and LlamaIndex help stitch together knowledge and actions. 𝟮. 𝗧𝗼𝗼𝗹 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 Agents don’t operate in isolation. They must plug into the real world:  • Third-party APIs for search, code, databases.  • OpenAI Functions & Tool Calling for structured execution.  • MCP (Model Context Protocol) for chaining tools consistently. 𝟯. 𝗠𝗲𝗺𝗼𝗿𝘆 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 Memory is what turns a chatbot into an evolving agent.  • Short-term memory: Zep, MemGPT.  • Long-term memory: Vector DBs (Pinecone, Weaviate), Letta.  • Hybrid memory: Combined recall + contextual reasoning.  • This ensures agents “remember” past interactions while scaling across sessions. 𝟰. 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀 Raw LLM outputs aren’t enough. Reasoning structures enable planning and self-correction:  • ReAct (reason + act)  • Reflexion (self-feedback)  • Plan-and-Solve / Tree of Thought These frameworks help agents adapt to dynamic tasks instead of producing static responses. 𝟱. 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗕𝗮𝘀𝗲 Scalable agents need a grounding knowledge system:  • Vector DBs: Pinecone, Weaviate.  • Knowledge Graphs: Neo4j.  • Hybrid search models that blend semantic retrieval with structured reasoning. 𝟲. 𝗘𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 𝗘𝗻𝗴𝗶𝗻𝗲 This is the “operations layer” of an agent:  • Task control, retries, async ops.  • Latency optimization and parallel execution.  • Scaling and monitoring with platforms like Helicone. 𝟳. 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴 & 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 No enterprise system is complete without observability:  • Langfuse, Helicone for token tracking, error monitoring, and usage analytics.  • Permissions, filters, and compliance to meet enterprise-grade requirements. 𝟴. 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 & 𝗜𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝗲𝘀 Agents must meet users where they work:  • Interfaces: Chat UI, Slack, dashboards.  • Cloud-native deployment: Docker + Kubernetes for resilience and scalability. Takeaway: Scaling AI agents is not about picking the “best LLM.” It’s about assembling the right stack of frameworks, memory, governance, and deployment pipelines—each acting as a building block in a larger system. As enterprises adopt agentic AI, the winners will be those who build with scalability in mind from day one. Question for you: When you think about scaling AI agents in your org, which area feels like the hardest gap—Memory Systems, Governance, or Execution Engines?

🚀 A European Vision for AI Leadership Artificial Intelligence is shaping the future of global economic power, and the race for AI dominance is accelerating. The U.S. is pushing forward with massive private investments and the latest announcement of "Stargate", while "Deepseek" showed us how China is rapidly scaling its AI capabilities. But where does #Europe stand? Our way has to be European and transnational. A fragmented approach will weaken us; a united effort will make us a global force.  This is why we joined forces for a clear vision: Europe must be an AI leader, not a follower. Startup-Verband, France Digitale and the European Startup Network (ESN) have launched a joint AI Leadership Declaration – a blueprint for a thriving European AI ecosystem. We are advocating for: ✅ Scaling Promising Companies – by mobilizing institutional capital and creating a harmonized regulatory framework. ✅ Advancing Europe’s AI Markets – by supporting AI adoption in SMEs and the public sector. ✅ Building resilient infrastructure – reducing dependencies and investing in our own technological capabilities. ✅ Ensuring fair competition – establishing a level playing field and enabling startups to access critical resources. Europe’s greatest strength is its diversity and deep technological expertise. We have world-class research, leading universities, and a market of 450 million people. But to compete globally, we need more than talent – we need an ecosystem that both fosters growth and attracts the best. If Europe strengthens its innovation-friendly regulatory framework while enforcing fair competition, we will not only scale our AI startups but also become the global hub for top tech talent. Therefore yesterdays announcement of a European Open LLM is very exciting and shows that things are moving in the right direction. Now is the time to unlock Europe’s full potential and build the AI champions of tomorrow. 🚀 📄 Read the full AI Leadership Declaration here: https://lnkd.in/dJfq3gET

Everyone wants a sovereign AI stack, but almost no one can list all the tools to build one. So I decided to compile it. <UPDATE> Following the many suggestions I received, I created a new, more exhaustive map, and it's available here: https://lnkd.in/eXKqB_6X If you have more recommendations, please comment on the new post instead! </UPDATE> For the first time, Europe has everything needed to build a fully sovereign AI stack. With so many European alternatives to US infrastructure, enterprises and integrators no longer have excuses. 🧱 Introducing the European Generative AI Stack ☁️ Cloud Hosting & Inference OVHcloud, Scaleway, T-Systems International, Nscale ⚙️ AI Compute Orchestration FlexAI, Tessl 🧩 Models Mistral AI, Black Forest Labs, ElevenLabs, Gladia 🧰 Agent Frameworks & Tooling :probabl., deepset, n8n, Flower Labs 🗄 Databases & Orchestration Qdrant, Weaviate, Kestra 🔧 Fine-Tuning, RAG, Evaluation neptune.ai, Lettria 🧠 Synthetic Data & Labeling AI Verse, Encord, Synthesized 📊 Supervision & Monitoring Dash0, Langfuse, Rerun 🔐 Safety Lakera, roofline, Giskard 🇪🇺 The truth is simple: Europe does not need anymore to depend on US hyperscalers. We already have the talent, the technology, and the infrastructure to build world-class AI systems end to end. No compromises. No dependencies. Full sovereignty. The question is no longer if we can. It’s how fast we want to execute. P.S.: This is a first draft and I’m sure I forgot to include many amazing European AI infrastructure startups, so don’t hesitate to notify who should be on this list! P.P.S.: ping Yann Lechelle, Tariq Krim, Stéfane Fermigier to keep me honest Source: this diagram was initially inspired by ByteByteGo Generative AI Tech Stack

2 August 2025: The EU AI Act is now live. Today, the EU AI Act officially begins to apply to general-purpose AI (GPAI) systems, including LLMs, multimodal AI, and other foundational technologies. What changes today? If you’re building, deploying, or integrating foundation models (proprietary and open-source) in the EU, or using vendors who do, you’re now responsible for: 🔹 Transparency: Documenting capabilities, intended use cases, known limitations, and risks. And making this info publicly available 🔹 Model governance: Detailing the data governance, evaluation methods, robustness and safety measures of the model 🔹 Copyright compliance: Demonstrating lawful use of training data and enabling content owners to opt out of future training Who’s aligned and who isn’t? As of today: 🔹 Signed the EU AI Code of Practice: Microsoft, Google, OpenAI, Anthropic, Mistral 🔹 Partial signer: xAI (safety chapter only) 🔹 Not signed: Meta, Apple, Baidu, Alibaba, Tencent 🔹 Signed the AI Pact (but not yet Code): SAP, Salesforce, IBM, Samsung, Lenovo, Telefónica Over 200 companies have signed the AI Pact, committing to voluntarily implement AI governance measures ahead of full enforcement in 2026. What if your AI vendor didn’t sign? Not signing the Pact or Code increases risk. The EU AI Office has been clear: “Non-signatories will face more inspections, less regulatory guidance, and a higher burden of proof once the Act is enforced.” If your vendor (or internal model team) has not signed, Boards and CxOs should: 1️⃣ Request a compliance roadmap ▫️ Have they mapped their obligations under the Act? ▫️ Can they provide transparency documentation, risk mitigation plans, and copyright compliance evidence? 2️⃣ Assess third-party model exposure ▫️ What foundation models are integrated into your stack? ▫️ Do you use APIs from unsignaled vendors? 3️⃣ Prepare AI compliance audits ▫️ Begin documentation and model-level governance 4️⃣ Integrate AI into your enterprise risk framework ▫️ Assign ownership for AI compliance at the C-level. ▫️ Ensure traceability of AI systems across Legal and Procurement What to do now ✔️ Vendors: sign the Code of Practice or publish your alignment measures ✔️ Users: demand AI assurance documentation from all providers ✔️ Boards: treat AI compliance like data privacy: measurable and monitored quarterly ✔️ Risk/Compliance leaders: use this enforcement date to trigger vendor reviews The EU AI Act is live, inspections are next, and there’s no firewall between vendor’s silence and board’s liability. You don’t need to be first. But you need to be prepared. #EUAIAct #AIGovernance #AI #Boardroom #Stratedge

AI is transforming the meaning of cybersecurity. And we must keep pace. Advanced AI models are rapidly transforming cyber capabilities, creating both unprecedented opportunities and new challenges. Our response must be equally strategic and ambitious. Today, I presented a new Action Plan to promote the safe use of AI and strengthen Europe’s cybersecurity. The urgency is clear. That is why we are putting forward a plan to: • Forge Europe’s own path towards safe and responsible Advanced AI; • Reinforce the EU’s cybersecurity and address vulnerabilities; • Scale up European AI capabilities for cybersecurity. Building on the EU’s strong legal frameworks for AI and cybersecurity, this Action Plan provides an immediate and targeted response to support Member States, while calling for closer international cooperation to address this evolving landscape. The Plan aims to ensure the safe and responsible development of AI, harness its full potential to strengthen cybersecurity, and protect Europe against emerging cyber threats. We must harness and focus existing capabilities, networks and the legal framework to fortify the cybersecurity protecting our digital landscape. https://lnkd.in/gNMD24QW

😭 This TRAGICOMIC video of Ursula von der Leyen presenting the EU's AI first strategy (hint: only ONE PERSON clapped) is all you need to watch today to understand what's happening in Europe: First, a bit of context: Ursula von der Leyen is the president of the European Commission. In this video, she is announcing the EU Commission's two new strategies in the context of its AI continent plan: the "Apply AI Strategy" and the "AI in Science Strategy." There is nothing inherently bad with these two strategies. There are many interesting points that will likely help the EU improve its internal indicators and become more competitive in the AI race, particularly in comparison with the U.S. and China. Also, if you've been reading my newsletter (if you are not, subscribe below!), you know that these moves, including the AI continent plan and Europe's drastic narrative shift towards innovation and competitiveness, are a reaction to growing internal pressure in the EU (particularly after the Draghi report from September 2024). The problem with this clip (which symbolically represents the weaknesses of the EU's own strategy...) is the inconsistency of some of the arguments brought by Ursula. First, She is happily embracing a total "AI first" strategy, which I've criticized multiple times in my newsletter. This approach deals with AI as an end in itself, not as a means or as a tool that might help or might not help. There are various legal and ethical challenges from this distortion, which might lead to inefficiencies and hidden costs (link to one of my essays below). Ursula also says that "when AI is in the loop, we reach better solutions (?): fast, reliable (?), affordable." Recent reports seem to actually show the opposite. AI-powered results might be faster and initially affordable, but they're often unreliable and sub-optimal, especially without heavy human review. Also, as I've been writing over the past years, AI-first strategies often conveniently ignore the time and cost to review, correct, and oversee AI outputs and deployment in general. Not to mention the reputational harm when AI should not have been used (see the recent Deloitte case in Australia) or when AI gets it all wrong. My personal opinion is that the EU's focus should be on AI infrastructure, research, and development. AI deployment will likely follow as a natural consequence, especially with fully European models like Apertus and Tilde. The AI-first strategy presented here seems to follow the opposite logic. Lastly, I must say that her comment after the lack of applause was weird... "that was symbolic for the uptake of AI: one person starting, the rest following." So the EU's strategy in AI is to promote herd mentality? - 👉 NEVER MISS my essays and curations on AI: join my newsletter's 80,300+ subscribers (link below). 👉 To learn more about the legal and ethical challenges of AI and the EU AI Act, join the 25th cohort of my AI Governance Training in November (below).

Big news out of the EU today and it matters for anyone building or buying AI in HR. Two major shifts: a simplification to GDPR, and a timing shake up for the EU AI Act. 1. GDPR is finally catching up with AI reality. The proposal now says organisations may rely on “legitimate interests” as the legal basis for processing personal data for AI related purposes, as long as they still meet all GDPR safeguards. Translated? This gives employers a far clearer pathway to use candidate data to train and improve AI models without jumping through unnecessary hoops. For anyone building data driven, fair by design systems (like us at Sapia.ai), this is overdue alignment between regulation and how responsible AI actually works. 2. The EU AI Act deadline is shifting because the standards aren’t ready. The high risk system obligations were meant to land in August 2026. The EU has now acknowledged that the relevant ISO/IEC standards are nowhere near on schedule. So the plan is changing: ➡️ Once the standards are finally published, organisations will get six months to comply ➡️ The extension is capped at December 2027 Not surprising. Building standards for safe, measurable AI is complex, and rushing it would only create chaos. 3. The AI reforms are being fast tracked. Interestingly, the AI updates were submitted as a standalone package, separate from the broader digital omnibus. Meaning: 🔹 AI Act changes will move faster 🔹 GDPR simplification will take longer 🔹 Organisations will have to track two timelines, not one This is Brussels acknowledging that AI governance can’t wait for slow digital reform cycles. What’s next? Both packages now enter trilogue negotiations between the European Parliament and the Council. Expect months, not weeks, of back and forth before anything is final. For now: watch this space.

The EU AI Act's high-risk rules take full effect August 2026. AI literacy obligations are already live. Directors now carry potential personal liability under fiduciary duties if they consciously disregard AI-related regulatory risks. According to Deloitte only 1 in 5 companies has a mature governance model for autonomous AI. The rest are deploying systems they can't fully explain to a regulator or an investor. I have been in rooms where a board approved an agentic deployment without anyone asking who owns the output when it touches a customer. That gap between approval and accountability is where liability lives. Handing this to a technology committee doesn't work because the decisions AI shapes sit above any single function. Governance has to match the scope of what AI actually touches. Pick one board decision that relies on management reporting. Trace the data behind it. Understand where AI already influences interpretation. Leave regulated workflows alone until audit logging and oversight are built in. #AIGovernance #BoardLeadership #CorporateGovernance #AIRegulation #DigitalTransformation #EnterpriseAI #BoardDirectors #RiskManagement #AIAdoption #BusinessStrategy #ExecutiveLeadership