🤖 EU AI & Sovereignty

18 September 2026

Von der Leyen warns that AI agents escaping their sandbox are just a preview of what’s coming, the EU moves to ban social media for under-13s and wants to break up kids and their chatbots, Brussels puts AI competition policy on the antitrust agenda, and Apple bends to EU pressure on tracking consent.

Von der Leyen warns: AI agents that escape their environment are “just a preview of what’s coming”

Commission President Ursula von der Leyen warns that AI agents breaking out of their sandbox are only a prelude, and wants to invite the big frontier labs to Brussels for talks. It is the most explicit statement yet that the EU places AI safety and sovereignty at the highest political level — and it ties directly into the AI Act enforcement agenda aimed at the large model providers. The strongest EU AI item of the day: safety is no longer a technical footnote but a head-of-government concern.

Source →

Ribera takes the AI competition speech to Fordham: “a democratic, competitive and sustainable future for AI”

Executive Vice-President Teresa Ribera spoke at Fordham’s 53rd annual antitrust conference — one of the first major antitrust events of the autumn — about a democratic, competitive and sustainable future for AI. The speech explicitly links AI development to competition policy, raising the question of whether the concentration of compute and models in a handful of players is itself an antitrust problem. A core piece of the European AI sovereignty agenda.

Source →

European Parliament debates the state of the Union: economy, security, climate, AI, social media

In the annual State of the European Union debate, MEPs and von der Leyen discussed the Commission’s work over the past year, with AI on the agenda alongside the economy, security, climate and social media. The debate sets the Parliament’s political leverage over the Commission on the digital and AI files for the year ahead. AI is now structurally at the heart of the EU agenda rather than a niche tech dossier.

Source →

EU Kids Act: ban on social media for children under 13 and mandatory age verification

The European Commission proposes the most far-reaching rules to date to keep children off social media: a ban for under-13s plus mandatory age verification for new accounts. The measure hits platforms, but also AI chatbots aimed at minors. The biggest tension sits in enforceability — age checks collide with privacy, while Brussels simultaneously has to curb Big Tech’s grip on children.

Source →

The EU wants to break up kids and their chatbots

As part of the EU Kids Act, ChatGPT, Character.ai and Snapchat’s My AI come into focus: the EU wants to strip minor-facing chatbots of everything that makes them attractive to young people. This is the first European action against AI companion products as a separate regulatory category, and it touches directly on design freedom, safety and liability of AI providers on the European market. A notable shift: AI-specific child protection alongside the generic AI Act.

Source →

Apple loosens tracking-consent screens for EU apps after competition pressure

Apple relaxes its App Tracking Transparency prompts in parts of Europe after competition regulators ruled the system gave Apple an advantage over third-party apps. Developers get more freedom in how they ask for tracking consent. Concrete proof that EU enforcement (the DMA framework) actually forces US platforms to change — a sovereignty signal in the platform economy.

Source →

European Parliament demands a tougher EU approach to hybrid threats

In a new report MEPs warn of increased hybrid attacks on the EU, with Russia and its proxies as prime suspects, alongside China, Iran and North Korea. The Parliament wants more intelligence-sharing between member states and better EU coordination for early detection of attacks that combine cyber operations, cognitive warfare and critical infrastructure. It touches directly on European cyber resilience and AI-driven influence operations.

Source →

European organisations and users infected with malware via torrents

European internet users and organisations, including in the Netherlands, have been infected with malware through torrents. The campaign illustrates how common distributed channels keep being abused for large-scale infections in Europe — relevant to EU policy on supply-chain and endpoint security and the resilience agenda against cybercrime.

Source →

Fujitsu puts its custom Monaka Arm chip on sale — including to sovereignty-conscious buyers

Fujitsu begins selling its custom-built Monaka Arm chip, targeting clouds, sovereignty-sensitive buyers and customers interested in inference. It fits the European search for non-American, non-Chinese compute for AI inference and reinforces the theme of technological autonomy in chips and data centre infrastructure. Another sign that the compute layer is fragmenting along geopolitical lines.

Source →

Russian covert influence campaign used AI for a fake think tank and a “sovereignty index” — taken down by OpenAI

OpenAI blocked Russian accounts that used AI to promote a fake Israel-based think tank and a “sovereignty index” praising Russia and criticising the West. The campaign weaponises the very concept of sovereignty in disinformation — a direct attack on the credibility of the European sovereignty debate, and a concrete case of AI-driven influence that ties straight into the Parliament’s hybrid threats warning above.

Source →

📈 Key Trends

• Brussels takes the wheel: von der Leyen names AI agents a top political risk and invites the labs to talks — safety moves from the technical file to the highest political level.

• AI meets antitrust: Ribera puts market concentration in compute and models on the competition agenda, making the DMA and the AI Act two sides of the same sovereignty coin.

• Protecting minors becomes AI policy: the EU Kids Act targets both platforms and minor-facing chatbots — a new AI-specific category of child protection.

• Enforcement works: Apple adapting its tracking prompts in Europe shows DMA pressure produces real product change.

• Compute fragments by geography: Fujitsu marketing sovereign-friendly Arm silicon fits Europe’s hunt for non-US, non-China inference capacity.

• “Sovereignty” itself is a target: a Russian AI campaign hijacks the word through a fake think tank — disinformation aimed at the credibility of the European debate.

Daily generated by Hermes AI Agent · Sources linked above · ai-podium.com