Sources: the European parliament is close to finalizing tough new measures on AI, including forcing chatbot makers to reveal if they use copyrighted material
Context & Ripple Effects
This Financial Times report caught the European Parliament mid-negotiation on what became the AI Act, and the arc since has run fast: weeks later lawmakers pushed forward a draft explicitly requiring companies deploying generative AI to disclose copyrighted material used for training, then the Parliament passed the full Act in June.
What remains is ratification — both the Parliament and the EU's 27 member states must approve the final text, which carries fines of up to €35M or 7% of global turnover for violations. The copyright-disclosure measure flagged here is the piece most directly aimed at chatbot makers.
First-order effects
- Chatbot and generative-AI providers serving EU users would have to reveal whether their models were trained on copyrighted material — turning training-data provenance from trade secret into disclosed fact.
Second-order effects
- Rights holders gain documented leverage to demand licensing terms once usage is disclosed, and any company deploying generative AI tools in the EU inherits transparency obligations regardless of where the model was built.
Third-order effects
- With penalties scaled to global turnover, the Act positions disclosure as an export requirement: model developers outside the bloc face a choice between restructuring training-data practices for EU access or ceding that market.
The trend: AI regulation is consolidating around mandatory training-data disclosure backed by turnover-scaled fines, with the EU's Act emerging as the template other jurisdictions will weigh.