European Commission Establishes AI Act Oversight Bodies as Ireland Takes EU Presidency
EC announces Advisory Forum and Scientific Panel members to oversee AI Act compliance, as Ireland assumes EU presidency with AI summit planned for October.
European Commission Forms Key AI Act Oversight Bodies
The European Commission has announced the members of two critical oversight bodies established under the AI Act: the Advisory Forum and the Scientific Panel.
The Scientific Panel comprises 60 members and is expected to act independently while playing an active role in the oversight of General-Purpose AI (GPAI) models. The Advisory Forum currently has 172 members and was conceived as a bipartisan forum of technical experts with balanced representation of commercial and non-commercial representation.
Code of Practice on AI-Generated Content Published
The European Commission published the final Code of Practice on marking and labelling of AI-generated content, which operationalises the AI Act’s transparency requirements. OpenAI has been the first actor to publicly announce that it will sign the Code of Practice on marking and labelling of AI-generated content.
OpenAI also published its Frontier Governance Framework, which serves as its Safety and Security Framework under the General Purpose AI Code of Practice.
Ireland Takes EU Presidency with AI Focus
Starting July 1st, Ireland will take over the Presidency of the Council of the European Union after Cyprus. Ireland’s presidency programme aims to strengthen the EU’s capacities in cloud computing and artificial intelligence, and announces a forthcoming AI Summit to take place in Dublin on 14 October 2026.
European Tech Sovereignty Push
The European Commission published its Tech Sovereignty Package, headlined by two policy documents and two legislative initiatives: the Cloud and AI Development Act (CADA) and the Chips Act 2.0. The CADA proposes to triple the EU’s data center capacity within the next 5-7 years, including by speeding up environmental assessments.
National AI Initiatives and Export Controls
Germany announced the establishment of an AI Security Institute. However, the EU faces challenges on the global stage: Anthropic disabled foreign access to the Mythos model after an export control directive by the US government which suspended access to Anthropic frontier models Fable 5 and Mythos 5, effectively shutting down European Union access to these models.
Earlier, Anthropic had expanded the range of actors with access to its Mythos frontier model under Project Glasswing to include eight EU countries, including Germany.
Liability Ruling on AI Tools
A regional court in Germany ruled that Google can be held directly liable for incorrect answers created by its AI overview tool.
Developments since publication
-
A paper titled 'SAOT: Self-Supervised Continual Graph Learning with Structure-Aware Optimal Transport' (arXiv:2607.00377) was accepted by the ICML 2026 conference. Source
-
A paper titled 'Watermarking for Proprietary Dataset Protection' (arXiv:2607.00325) was presented at the ICML 2026 Workshop on Trustworthy AI for Good. Source
-
A paper titled 'MolSafeEval: A Benchmark for Uncovering Safety Risks in AI-Generated Molecules' (arXiv:2607.00464) was accepted by Findings of ACL 2026. Source
-
A paper titled 'Device Passport: Enabling Spatio-Temporal Pretrained Models to Generalize Across Input Layouts' (arXiv:2607.00249) was presented at the ICML 2026 Workshop on Structured Data for Health Source
-
A paper titled 'AdaBoosting Text Prompts for Vision-Language Models' (arXiv:2607.00684) was accepted to ECCV 2026. Source
-
Ireland secured €10 million in joint European and Irish Government funding to establish the AIF IRL-Antenna. Source
-
The initiative is supported by the Irish Government through the Department of Further and Higher Education, Research, Innovation and Science (DFHERIS), alongside European funding provided through Euro Source
-
Google DeepMind's mathematical reasoning system scored in the top 1% on International Mathematical Olympiad (IMO) problems. Source
-
Google DeepMind's mathematical reasoning research team published their approach in a paper submitted to NeurIPS 2026. Source
-
Research groups at University of Cambridge and UCSF published separate but complementary approaches in July using diffusion models adapted for molecular dynamics to predict protein dynamics. Source
-
Research at MIT and Stanford analyzed reasoning models and found that the key factor for success on hard mathematical and logical problems is not model size but how the model is trained to self-correc Source
-
DeepMind published work on a new training approach called 'prospective credit assignment' for teaching models to anticipate how current decisions will affect outcomes many steps ahead. Source
-
DeepMind's prospective credit assignment approach showed meaningful improvement in success rates on SWE-Bench for issues requiring more than 10 steps to resolve. Source
-
Research by the Allen Institute for AI found that hallucinations in language models are more likely to occur when a model is asked about facts that were underrepresented in training data. Source
-
Meta AI released results on video understanding models with the ability to answer questions about what happens across long video sequences (30+ minutes). Source
-
Research published at ICML 2026 described a new training method called 'selective activation sparsity' that trains models to use only the most relevant parameters for each specific task. Source
-
Models trained with selective activation sparsity performed comparably to models three times their size on reasoning benchmarks. Source
-
Ireland's publicly released programme for the EU presidency aims to strengthen the EU's capacities in cloud computing and artificial intelligence. Source
-
The Digital Omnibus proposal is a draft law seeking to amend various digital laws including the General Data Protection Regulation. Source
-
The German court argued that contrary to the traditional liability exemption applicable to search engine operators when presenting third-party content, the AI overview created new and independent cont Source
-
The German court rejected the argument that general awareness of the unreliability of AI-generated content protected Google from liability. Source
-
Under its obligation to identify systemic risks arising from its models, OpenAI lists four compulsory risk categories under the Code of Practice: cyber offence, chemical/biological/radiological/nuclea Source
-
OpenAI does not identify any further systemic risks beyond the four compulsory risk categories, such as risks related to fundamental rights, in its Frontier Governance Framework. Source
-
Everhaze raised €450k from three Irish family offices (Astogo Holdings, Kearns family office, and hotelier Mark Cosgrave) on the eve of launching its conversational agentic AI product Lú across Irelan Source
-
The three Irish family offices backing Everhaze have a combined net worth of more than €350m. Source
-
The PRII Census 2025 showed that 83% of PR professionals are working beyond their contracted hours, putting in an average of almost 10 extra hours per week. Source
-
The Irish in-house public-sector share of the PR workforce grew to 42% in 2025, up from 25% in 2019. Source
-
Everhaze has identified 48 separate tasks that Lú can carry out, including text editing, media list management, media monitoring and report generation. Source
-
The AIF IRL-Antenna partnership involves seven regional innovation hubs in Ireland. Source
-
The antenna is described as Ireland's national programme connecting the Irish AI ecosystem to European AI Factories infrastructure under the European High Performance Computing Joint Undertaking (Euro Source
-
The Regulation of Artificial Intelligence Bill 2026 establishes the AI Office of Ireland as an independent statutory body to act as Ireland's central coordinating authority for the implementation of t Source
-
The Bill is structured in 10 Parts with 139 Sections and 4 Schedules. Source
-
The Bill provides Market Surveillance Authorities with a structured, proportionate enforcement toolkit progressing from cooperative compliance (notices) to coercive measures (prohibition, seizure) and Source
-
A major area of focus for Ireland's presidency will be the Digital Omnibus proposal, a draft law seeking to amend various digital laws including the General Data Protection Regulation, after member st Source
-
Ireland has launched the AI Factory Antenna (AIF IRL-Antenna) with €10m in joint European and Irish Government funding. Source
-
The AI Factory Antenna will be operated by the Irish Centre for High-End Computing (ICHEC) at University of Galway, in partnership with CeADAR, Ireland's Centre for AI. Source
-
The antenna aims to grow adoption of and innovation through AI in the country while securing a foothold within the wider European AI ecosystem as part of a network of AI Factories and high-performance Source
-
The antenna operates under the European High Performance Computing Joint Undertaking (EuroHPC JU), providing AI-optimised high-performance computing. Source
-
The partnership involves seven regional innovation hubs to ensure that all companies can equally benefit, irrespective of their location. Source
-
The European Commission published its Tech Sovereignty Package, including the Communication on European Tech Sovereignty and EU Open Source Strategy, a Strategic Roadmap for Digitalisation and AI in E Source
-
The Cloud and AI Development Act (CADA) proposes to triple the EU's data center capacity within the next 5-7 years. Source
-
Ireland's presidency programme announces a forthcoming AI Summit in Dublin on 14 October 2026, bringing together EU and global leaders, heads of government, CEOs, investors, innovators and academics. Source
-
A major area of focus for Ireland's presidency will be the Digital Omnibus proposal after member states failed to agree on a common position during negotiations under the Cypriot presidency. Source
-
Cate Blanchett unveiled the 'Human Consent Registry', a tool that would enable artists to provide or withhold permission for their image to be used by artificial intelligence. Source
-
Norway moved to restrict the use of generative AI by primary school students in the classroom, following a documented drop in students' basic skills with one in four Norwegian students reading below t Source