Blog
-
AI advice suppresses people’s willingness to say “I don’t know”, even when the advice is wrong and accuracy is incentivizedAI systems can reduce a person’s willingness to admit they do not know an answer. -
Analysis of the integration of Artificial Intelligence in VET systemsThe European Commission’s Joint Research Centre recently analyzed how Artificial Intelligence is being integrated into Vocational Education and Training (VET... -
AutoSynthesis: An agentic system for automated meta-analysisWorth knowing that researchers have developed a multi-agent AI system to automate the process of scientific meta-analysis. -
Fable 5 is great, but you don't need it.Anthropic has released Fable 5, a model designed to act as an autonomous worker for complex tasks. -
✍️AI Office publishes frontier AI expert findings on EU competitiveness, sovereignty and securityHere is a summary of the latest findings from the EU AI Office on frontier AI and European competitiveness.
-
ARC-AGI-3: A New Challenge for Frontier Agentic IntelligenceResearchers have released ARC-AGI-3, a new benchmark designed to test how well AI systems can solve novel, abstract problems.
-
ARC-AGI-3 Leaderboard & Scores — July 2026New results from the ARC-AGI-3 leaderboard show how AI models perform when learning new rules through interaction.
-
✍️Directly Responsible Individuals (DRI)Identifying a “Directly Responsible Individual” (DRI) is a key step in managing the risks of autonomous AI.
-
Statistically Undetectable Backdoors in Deep Neural NetworksWorth knowing that backdoors in deep neural networks can remain hidden even when users have full access to a model’s weights.
-
✍️The Triad of Tech Sovereignty and Europe’s Package:The European Commission recently introduced a new Technological Sovereignty Package to strengthen digital autonomy across the continent.
-
Open Data Policy Lab: 100 examples of how non-traditional data serves the public interest.This collection includes data from satellites, connected devices, commercial transactions, and online communities. It covers specific use cases like detectin...
-
3 Levels of AI Use DisclosureMoving beyond simple “AI or not” checkboxes is necessary for effective governance. -
✍️EU Action Plan on Cybersecurity and Artificial IntelligenceThe European Union has released a new action plan to manage the intersection of artificial intelligence and cybersecurity.
-
✍️The European Data Protection Board is opening a consultation on new guidelines for data anonymisation.The board is seeking feedback on Guidelines 02/2026. These rules aim to clarify how personal data can be processed so that individuals are no longer identifi...
-
Il Digital Ominbus tra competitività economica e arretramento delle garanzie fondamentali - Nexa Center for Internet & SocietyThe European Commission recently proposed the Digital Omnibus to simplify and consolidate the EU’s digital rules. -
GitHub - sev7enITA/policywatcher: AI-powered regulatory intelligence platform for Big Tech and FinTech corporate policies. Monitors 16 companies with 15-KPI governance framework, bilingual EN/IT, Gemini 2.5 Flash analysis.Here is a project that monitors how big tech companies change their corporate policies using AI.
-
AI Safety Index — Summer 2026 | Future of Life InstituteThe Summer 2026 AI Safety Index provides a new look at how leading companies handle security and risk. -
AI and the Future of Publishing: Not Detection, but AnswerabilityIdentifying AI-generated content through detectors and watermarks could not be the right way to ensure trust or quality. -
No data, no AIAI success depends on the quality and structure of the underlying data. -
RL+LLM Wiki — Updated Report (Isometric Town Frame)A new project is using multi-agent systems to create a living wiki of reinforcement learning research. -
EU AI Act Pushes ISO/IEC 42001 Into AI Compliance PlanningThe ISO/IEC 42001 standard provides a framework for organizations to manage AI risks, documentation, and human oversight. It is the first global standard fo... -
cisco-ai-defense/defenseclawCisco released a framework called DefenseClaw to help manage security governance for Agentic AI. This framework addresses the transition from basic chatbots...
-
✍️Anthropic’s Economic PolicyAnthropic’s June 2026 framework addresses AI’s labor-market impact via a tiered, unemployment-triggered response structure, resting on three foundations — me...
-
The 6 security shifts AI teams can't ignore in 2026AI security is shifting as systems move from simple tools to autonomous agents capable of taking direct actions in enterprise environments. -
AI has become one of the most powerful gatekeepers in travel - Scottish Licensed Trade NewsAI platforms are becoming the primary gatekeepers for travel and hospitality recommendations. -
Current Advances in LLM Reasoning — ACL 2026 TutorialResearchers are developing new ways to measure and improve how Large Language Models (LLMs) reason through complex tasks. The LLM Reasoning Tutorial covers ... -
HKUDS/LightRAGLightRAG offers a way to make Retrieval-Augmented Generation (RAG) systems faster and less complex.
-
✍️Preliminary Report of the UN Independent International Scientific Panel on AIA new manual provides a clear summary of AI guidelines for public services.
-
Companies are scrambling to stop employees from maxing out AI budgets with small tasks | TechCrunchCompanies are beginning to limit how much AI tokens employees use for basic tasks. -
headroomlabs-ai/headroomMany AI systems waste resources by processing unnecessary data.
-
✍️Open Source AI Gap MapA new project has mapped the current state of the open-source AI ecosystem.
-
Technical Support Instrument - PerformanceThe European Commission uses the Technical Support Instrument to provide technical expertise to Member States facing complex reform challenges. -
microsoft/generative-ai-for-beginnersMicrosoft released a large-scale educational resource that teaches the fundamentals of generative AI.
-
Final training runs account for a minority of R&D compute spendingMost of the money spent on AI research goes toward experiments rather than the final model training. -
Distributed Attacks in Persistent-State AI ControlAI coding agents now work on persistent codebases, creating a new surface for distributed security attacks.
-
LACUNA: A Testbed for Evaluating Localization Precision for LLM UnlearningResearchers have developed a new way to test if AI models truly forget sensitive information.
-
Online Safety Monitoring for LLMsResearchers have found that alignment training is not enough to ensure AI safety during live use.
-
World Wide Models: Literary Tools for Cultural AILarge language models currently risk creating a monolingual AI landscape that overlooks cultural diversity.
-
AI usage patterns are shaped by perceived gains in human agencyPeople who use AI chatbots often report feeling more capable, even when the AI’s reliability is inconsistent.
-
AI and corporate responsibility – From fragmented compliance to unified governanceThe EU AI Act is shifting the focus from voluntary AI ethics to mandatory corporate governance. -
Great Tables `v0.22.0`: footnotes, summary rows, column merges, and moreThe Great Tables Python library recently released version 0.22.0 with several new features for data visualization. -
From dependency to resilienceThe EU is focusing on reducing its reliance on technology from outside the bloc. -
Software Engineers Are Facing an Existential Crisis As They Drown In Horrendous AI CodeSoftware engineers are finding it harder to manage the influx of low-quality code generated by AI. -
Open source (OSOR)OSOR is the European Commission’s platform where the Free and Open Source Software community publishes news, events, and solutions for public administrations... -
Nano Banana 2 LiteGoogle released Nano Banana 2 Lite, a new image model designed for speed and cost efficiency. -
Best Cases Award - 2026 EditionThe Public Sector Tech Watch has announced the 2026 Best Cases Award for European public administrations.
-
✍️www.nber.orgResearch shows that while AI significantly speeds up writing code, it does not always lead to faster software releases.
-
✍️CAISS analysis of Anthropics system card for Fable 5 and Mythos 5 | SimulaResearchers have analyzed the transparency and reporting practices in Anthropic’s latest AI system cards.
-
From fragmented to seamless: A framework for omnichannel public service delivery in the EUThe EU is developing a framework to make public services easier for citizens and businesses to use. -
“Doctor, it hurts when agents create unreviewable PRs.” “Don’t do that.”AI agents are producing large amounts of code that can be difficult for humans to review.
-
GitHub - baidu/Unlimited-OCR: Unlimited OCR Works: Welcome the Era of One-shot Long-horizon Parsing.Baidu released Unlimited-OCR to handle long-horizon document parsing.
-
Landscape of EU digital regulationsThe EU has established a comprehensive legal framework to govern data, privacy, and digital services.
-
AI Risk Management Newsletter #40New reports and consensus statements are providing more structure for AI risk management. -
The European Union is deploying AI across strategic sectorsThe European Union is moving from high-level AI policy to practical deployment across several key industries. -
The EU AI Act Newsletter #105: Transparency Tools LandThe EU AI Act is moving toward practical implementation with new transparency tools and updated timelines. -
Anthropic Economic Index report: CadencesHere is a look at how AI usage patterns align with human daily rhythms.
-
Cheshire Cat AICheshire Cat AI is a new framework designed for building and deploying custom AI agents. -
Published: 26 June 2026Research shows that high scores on medical benchmarks do not always mean an AI model is ready for real-world healthcare use. -
Commission reaches preliminary position that Amazon's and Microsoft's market leading cloud services should be designated under the DMAThe European Commission has reached a preliminary view that Amazon and Microsoft should be designated as gatekeepers for their cloud computing services under... -
The Generative AI Fizzle™AI market valuations may be entering a period of slow decline rather than a sudden crash. -
The Industry of Lies, or What Leaders Need to Know About AIAI adoption involves practical challenges that go beyond the technology itself. -
Press corner | European CommissionThe European Innovation Council has selected EQT to manage a new €5 billion fund for European deep-tech scaleups.
-
Best Cases Award - 2026 EditionThe Public Sector Tech Watch has announced the 2026 Best Cases Award for European public administrations.
-
Shocking number of teens say they’ve been targeted by AI nudes: ‘It is disturbingly easy’One-third of American teenagers faced AI-generated nude images in a recent survey. -
✍️The geography of artificial intelligence in European regions: innovation, exposure and useThe geography of artificial intelligence in European regions: innovation, exposure and use AI development and adoption vary across the European Union.
-
Why building an AI Toolbox for European public services mattersThe European Commission is launching an AI Toolbox to help public administrations move beyond just testing new technology. The toolbox provides a shared rep...
-
Intacture: data centre beneath the mountain officially openedA new data centre has opened inside a dolomite mine in Italy. The Intacture project in Val di Non is now fully operational. It was built by a public-private... -
Towards AI strategic adoption: Europe’s next step on AI in PAEurope is moving from testing AI tools to integrating them into the core functions of public administration.
-
The State of AI, 2026The AI industry is moving from a period of initial excitement to a phase of practical reality. -
✍️Domyn-Led EUROPA Consortium Wins EU Frontier AI Grand ChallengeThe EU Commission has selected the EUROPA consortium to build a new open-source frontier AI model.
-
Students Are Experiencing AI in Very Different Ways. Is That a Problem? - Education WeekThe “digital divide” isn’t just about who has a laptop anymore; it’s about who has the literacy to use AI effectively. -
✍️Prompt Injection as Role ConfusionModels often struggle to distinguish between official system instructions and user input.
-
The EU AI Act amendments explained (Part 2)The proposed expansion of sensitive data processing for bias mitigation in the EU AI Act presents a nuanced regulatory challenge that warrants a closer look. -
Critical Ignoring: The Essential Skill for Surviving the 2026 Information FloodThe erosion of human attention is no longer a sociological byproduct; it has become a structural risk to institutional competence. -
The UK will scan asylum-seekers’ faces for age checks—despite knowing the tech is flawedUsing flawed technology to decide the legal fate of a child is a line we should never cross. -
AI Risk Management Newsletter #39Recent developments show a clear focus on establishing safety standards and risk management for frontier AI. -
AI Is Unsecurable: Four Things You Need To Do NowThe myth of “AI guardrails” is officially dead, and honestly, I’m relieved we’re finally stopping the charade. -
Accenture: Then and now, and how it may signify things to comeThe hype cycle for enterprise AI is starting to hit a very uncomfortable wall of reality. -
The EU AI Act amendments explained (Part 1)The EU isn’t backing down on AI regulation; they are just refining the roadmap. -
The EU AI Act Newsletter #104: Experts Take Their SeatsWe are finally moving past the “what if” phase of the EU AI Act into the “how exactly” phase. -
✍️The Fable 5 Export Controls Harm US Cyber DefenseWe are making a massive mistake by conflating “fixing a bug” with “crafting an exploit.”
-
The Governance Implications of AI TokenomicsWe need to stop treating AI costs as a simple IT budget line item and start treating them as a core governance risk. -
Why AI hasn’t replaced software engineers, and won’tThe narrative that AI is about to trigger mass layoffs in software engineering is mostly “AI washing.” -
Maybe Section 230 doesn’t shield AI companies from liability, after allThe “safe harbor” for AI companies is starting to crumble, and frankly, it’s about time. -
Attacking & Threat Modeling The Agentic Top Ten: ASI04 - Agentic Supply Chain VulnerabilitiesThe “hopium” phase of AI security is officially over. We have to stop pretending that we can simply “patch” our way into a secure agentic ecosystem. -
Banning AI is not an effective AI policyBanning AI isn’t a security strategy; it’s just an invitation for people to work in the shadows. -
What ‘Magnifica Humanitas’ means for AI Governance ProfessionalsPope Leo XIV has released “Magnifica Humanitas,” a significant new encyclical on AI governance. -
Choosing to Stay HumanI’m starting to notice a growing “cognitive surrender” in our professional feeds. -
The EU AI Act Newsletter #102: Pressure Builds over Anthropic's MythosThe EU is refining the rules for AI systems and addressing new security risks. -
Memory Is a Feature. It Is Also an Attack SurfaceAI agents use persistent memory to provide better assistance, but this also creates a new security risk. -
✍️The AI risk repository: A meta-review, database, and taxonomy of risks from artificial intelligenceResearchers have developed a unified database to standardize how we describe and manage risks in artificial intelligence.
-
AI Makes Work More IntenseResearch shows that generative AI tools can increase employee workloads instead of reducing them.
