Other
AI: capabilities, regulation, labour
Updated 8h ago
2/5
Models are learning faster than the institutions behind them can adapt, with this thread tracking frontier-model capability jumps, the AI Act and its enforcement, labour-market impact, and infrastructure (chips, energy, water).
Intel's €5 billion investment in Ireland expands EU AI chip manufacturing capacity, and a US court ruling further clarifies copyright issues for AI training data.
State of play
The EU AI Office's first systemic-risk investigation under the AI Act continues, testing the new capability-based regulatory trigger. Enforcement capacity across the bloc remains uneven, with key deadlines for high-risk AI systems under consideration for further prolongation to 2027. Legal interpretations of AI training data usage are diverging, with a recent US court ruling establishing "transformative fair use" for training copyrighted material, while allowing trials for alleged use of pirated copies. This contrasts with the EU's more documentation-heavy approach under the AI Act's GPAI obligations, creating potential friction for global model developers.
Frontier Model Oversight
A US voluntary framework for frontier-model oversight is now operational, with OpenAI's decision to delay GPT-5.6's public launch marking its first concrete application. This allows US government scientists up to 30 days for pre-deployment security reviews, focusing on threats like sophisticated cyberattacks. Microsoft, Google, and xAI have agreed to provide advance access to US authorities. Intelligence agencies from the Five Eyes alliance have issued a public warning about the imminent emergence of AI models capable of destructive cyberattacks, urging accelerated model evaluations and cross-border coordination. China is considering curbs on overseas access to its top frontier models, with proposals for a tiered regime. The EU Commission has adopted a new Action Plan on Cybersecurity and AI, which includes a mechanism to evaluate frontier systems before they are placed on the single market, expected to be operational by 2027. European Commission President Ursula von der Leyen and former U.S. Secretary of State Hillary Clinton have endorsed the creation of a Youth AI Safety Institute focused on evaluating AI tools for risks to minors.
Infrastructure and Sovereignty
The European Commission's proposed 'made-in-Europe' tech sovereignty package is in the legislative process. It aims to reduce reliance on foreign cloud, AI, and chip providers by setting sovereignty criteria for contracts in sensitive sectors. The package also promises fast-track approvals for data centers using European chips, linking AI infrastructure to industrial policy. This push for technological autonomy unfolds alongside tightening global export controls on advanced AI chips. The US has tightened its AI chip export controls, extending the ban to subsidiaries of Chinese companies located outside China, which affects European supply chains and intensifies EU debates on industrial policy. Intel has announced a €5 billion investment to expand AI chip manufacturing at its Leixlip plant in Ireland, adding hundreds of jobs.
Key events
Full chronicle426
This week
- Intel announced a €5 billion investment to expand AI chip manufacturing in Ireland.
- A US court narrowed artists' copyright claims against OpenAI, upholding transformative fair use for training data.