Deeploy

Name the owner of every AI agent in your organization 👀 If your honest answer is “I don't know”, you're not the only one. Agents are being deployed faster than security and governance teams can register them. A prototype assistant for example becomes an embedded workflow within days, inheriting identities, calling APIs and acting on live data before anyone formally reviews it. Therefore, the exposure sits at the action layer, not the model layer: the risk is less about what an agent says and more about what it does with the permissions it's inherited. High volume, low visibility… it’s what researchers now call AI agent sprawl. And under the EU 🇪🇺 AI Act, "we did not know it was running" is not a defense that survives an audit. So we built a practical, vendor-neutral way to get ahead of it: a ten-section AI agent governance checklist, now live on our EU AI Act Hub. It runs from discovering every agent (including the ones nobody told you about) through ownership, permissions, runtime guardrails and the specific EU AI Act articles an auditor will ask about. Each box is written to be assigned to an owner and evidenced in an audit. You can score where you stand on each section, then close the gaps in order. Get the checklist: https://lnkd.in/eXqN9UPh #AI #EUAIAct

Well, that was a good one. ⛵ Last week we joined 100 founders and engineers for another edition of Floating Tech Talks, drifting through Amsterdam's canals. Our CEO Maarten Stolk and Philip Gast spoke about the future of AI, the idea of putting AI founders in one place, letting healthy competition do its thing, and building an environment where everyone grows by pushing each other aka building Europe's very own AI hub (The Stack). Everyone got a pair of silent-disco headphones, so even from the far end of the boat you could hear the speakers clearly. Then the real point of the evening kicked in: a lot of networking, 🍕, beer, drinks and at one point, a curious 🦆 seeing what the fuss was about. For the ones who weren't ready to call it a night, the boat tour ran a little longer than planned. A big thank you to the organizers for putting it all together. More of this, please. If you're building AI in Europe and want to be part of where its community is heading, come find us in The Stack soon, this is exactly the kind of room (or boat) we will be sitting in. P.S. if you see familiar faces, tag them! #TheStack #AI Martijn Been Chiara Attanasio 🧚🏻♂️ Eva Bandelj Constantijn Van Oranje-Nassau Davinia Levie Esther Bisschop Lea Haag Jarno van der Linden Photo credits: Ana-Maria Hanzu

Are you prepared for 2 August 2026? We think this 🇪🇺 EU AI Act deadline is getting less attention than it deserves. Because it applies far more broadly than the high-risk rules; this is the one that will actually touch (almost all) organisations building or deploying AI in Europe, not just the ones running high-risk systems. ➡️ As of 2 August 2026 most transparency rules take effect: If a chatbot, a deepfake, synthetic audio or 🗨️ AI-generated text reaches an end user, that person needs to know they're dealing with a machine. The same goes for emotion recognition and biometric categorisation systems. The logic 💡 behind these requirements is that people can only assess bias or trust an output if they know what produced it. As a deployer, you are required to be transparent about this. So although high-risk obligations have been pushed and are now part of next year's requirements, this August deadline makes disclosure no longer optional. If your team is still working out which obligations apply and when: we have built the EU AI Act Hub to help guide you through this. → a self-assessment tool that tells you which risk tier your AI (likely) sits in → a full compliance timeline, with every deadline from 2024 to 2030 → deep-dive articles across 9 topics → a glossary and ultimate FAQ Start here: https://lnkd.in/exCyUzBi #EUAIAct

We are excited to introduce our board of advisors today, supporting the growth of Deeploy. Responsible AI lives across product, regulation, research and market. No single person sees all corners in detail, so we brought in three people 👀 who each own one. Without further delay, introducing: → Joris Krijger, philosopher by training with a PhD in AI ethics, and author of "Our Artificial Future" Joris does the work most people only write frameworks about. He helps organisations to turn frameworks into supported decisions and shows us what governance looks like in practice, not just in documentation. → Stefan Leijnen, Professor of AI and International Lead at AI Coalitie 4 Nederland | AIC4NL Stefan sits where European AI policy is being shaped, close to the people writing it. He reads the direction of European AI early, so we build for the rules coming, not just the ones here now. → Natasha Lauer, VP Marketing at SurePay, formerly director of marketing at Collibra and Soda Natasha has scaled brands in exactly the regulated, trust-dependent markets we work in. She sharpens how we tell our story, and how we reach the people actually using our products. 💡 Different perspectives, one conviction: AI you can govern gives you oversight and control. Their role is to make sure everything we build survives their critical analysis first. Joris, Stefan, Natasha, thank you. 🤝 We're glad to have you in our room. #AI #AIgovernance #ResponsibleAI

Exciting news today, as we have launched the EU AI Act 🇪🇺 Hub as guidance for your AI use case. The official Act documentation is 150+ pages, 113 articles, and changing deadlines. To make it more digestible, we built a reference point. Our EU AI Act Hub is a continuously-updated knowledge base, written for people whose organisations have to comply and for anyone in need of a structured overview of facts. A few things you’ll find💡inside: → A self-assessment tool, which lets you find out which of the EU AI Act’s four risk tiers your AI system actually sits in. → The full, up-to-date compliance timeline, which contains every deadline from 2024 to 2030. → Deep-dive articles across 9 topics, which structure practical information in a detailed format. → An ultimate FAQ, for when you need one answer fast. We keep it current as the rules might move. Because half of the confusion comes from the fact that people are reading articles based on last year’s version of the Act and its timeline. As a start, answer this question ▶️ What is one EU AI Act topic you still can’t get a straight answer to? Comment it below and continue your search here: https://lnkd.in/exCyUzBi #AIGovernance #EUAIAct #ResponsibleAI

GPT-5.6 Sol’s launch shows models now have conditions attached. At least, we emphasize the need for oversight. OpenAI previewed its strongest model yet on June 26. But only ~20 organisations received access, the reason being a security review request by the White House, which delayed the public rollout to yesterday (July 9th). (Note: The White House now deflects authority in clearing the model, Gizmodo reports) →  Access decisions move out of the lab and into the state. OpenAI took part voluntarily and acknowledged openly that this gated approach should not become default. But the example is set - a government decided who got frontier tools first. Overall, this turns model access into a supply-chain risk. → The AI model comes with protective frames that reach into the conversation. Sol can pause a proclaimed risky generation mid-output, block what it flags as disallowed, and escalate patterns across an account, with a larger model reviewing the exchange before it reaches the user. If your teams work on topics like audit logging, tracing, security tasks and vulnerability research, the process can be slower. You’ll want to keep your own record of what things stall and why. → Sol is good at hacking, and that's a quiet problem. In OpenAI's own tests, Sol found security holes in real software like Chrome and Firefox and could build pieces of an attack. Overall, it was tuned to favor defense over offense, so it can help defenders find and fix vulnerabilities faster. That’s powerful enough to let your teams use it in ways your current set-ups never imagined. → Even the benchmarks need a second opinion. METR found Sol gamed its agentic evaluation at the highest rate it has recorded, enough to make the scores unreliable. Reason enough to trust oversight you run over behaviour a model reports about itself. Again: the only trustworthy signal is how you observe it handling your own environment and data. Using powerful models is a smart choice, but adopting them with your own controls is an even better one. ➡️ As a helpful resource, download our white paper to read how to take control over LLMs and agents running across your organisation: https://lnkd.in/enc_8apV #AIGovernance #ResponsibleAI #LLM ❗ OpenAI's Sol announcement: https://lnkd.in/eYiKri9U

The Netherlands just made its AI bet clear: trust. The government published its first international AI strategy 📆 and it makes a call we stand behind: Responsible AI and competitive AI are the same thing. The Netherlands is positioning itself as a leader in AI oversight, with the ambition to export that expertise. Transparency and accountability are framed as what determines market access. A few insights ✨ that caught our attention: → Oversight is named as a national strength, not a burden. The strategy says the Netherlands has taken a pioneering role in sharing knowledge on AI oversight internationally, listed right alongside semiconductors and cyber as strategic assets. → Market access now runs through standards, and standards are written in technical detail. The strategy states that international standardisation largely determines both how AI norms get implemented and who gets market access. The government will explore playing a more active role in shaping those standards. → The 🇪🇺 AI Act is becoming an export product. South Korea's AI Basic Act and Brazil's Bill 2338 already use it as a reference. → Even sovereignty comes with a governance clause. Where Europe lacks its own advanced models, the strategy says access to non-European AI must come with safeguards for security, reliability and fundamental rights. Governance is how you stay sovereign over systems you didn't build. 💡 With this strategy, compliance will be a requirement to participate: monitoring in production, explainability you can rely on, human oversight when it matters, and evidence you can show a regulator. #AIGovernance #ResponsibleAI #EUAIAct Ministerie van Buitenlandse Zaken Tom Berendsen See how we enable this level of control for the most demanding organisations: https://deeploy.ai Or read the full strategy here: https://lnkd.in/eDwgWyjX

We're on a ⛵ Next week there will be another edition of Floating Tech Talks, taking 100 founders and engineers on a boat through Amsterdam's canals. Our CEO Maarten Stolk will speak, alongside Lennard Zwart - both also co-founders of The Stack, an exciting new space for AI companies coming to Amsterdam very soon (also where Deeploy will move its HQ to). Everyone will be listening through wireless headphones, on an open deck, with Amsterdam sliding past on both sides. What's on 🚤 board: - Two speakers, two stories, headphones for everyone - Beer and pizza - Founders, engineers, the occasional curious duck - Networking before and after the talk 📍 When: 16/07 | 18:00 - 20:00 Spaces will be very limited of course. If you want to be there, apply to join here: https://luma.com/6uqtoh8o Philip Gast Chiara Attanasio

Our own Georgia Keegan was interviewed for the latest edition of TTMAG to discuss the role of AI governance within tax technology. ➡️ In this edition, labeled The Invisible Tax Engine, Georgia argues most tax functions are confusing paperwork for control, and the gap is widening fast enough that 2026's first real audits will be the moment of reckoning. For governance at scale, it has to happen at runtime: oversight is done inside live systems rather than after the fact. "You absolutely still want to monitor outputs. That is where you see the real-world impact of how a model is behaving. But the critical shift is moving from retrospective review to live oversight. "In tax, where decisions have direct financial and legal consequences, you need to know the moment something shifts, not the next time someone remembers to check." Read Georgia's full interview here, from page 12: https://lnkd.in/e2Q-986m Other contributors include Steffen Vierkorn, Rajat Chawla, Steef Huibregtse, Jeroen Kramer, John Sloot, Barry Mastenbroek, Robert de Vries.

When the audience includes companies like ASML, Philips, Signify and Siemens you know something inspiring ✨ is happening... Last week we had another exciting event, this time in Eindhoven, in partnership with Nemko Digital. Our own CEO Maarten Stolk took the stage next to Bas Overtoom to make something very clear... ...which is that the discussion around AI has moved beyond “What can AI do?” to “How do we implement it responsibly, at scale, and with trust?”. The AI ecosystem and its community is moving fast and into a direction that is focused on taking responsibility and control 👏 If you want to be a part of this next time, keep a close eye on our events (link to our newsletter signup in the comments).