EU AI copyright liability risks surge as fair use protections vanish

The absence of fair use in Europe means streaming and media companies have stronger leverage to protect their catalogs from unauthorized ingestion by models like Llama and Gemini. This structural legal difference forces AI developers to shift from mass scraping to formal licensing agreements to avoid massive fines under the EU AI Act. As the Netherlands emerges as a hub for collective redress, the industry must prepare for a wave of class-action litigation that could reset the valuation of training data globally. Watch for the CJEU ruling in Like Company v. Google in 2027 to establish the final boundary for generative AI data mining.

The EU AI copyright liability landscape is intensifying as rights holders across publishing, music, and media sectors file coordinated claims against major AI developers. In June 2026, a cluster of announcements signaled a real shift from AI research to commercial AI-driven network automation, but the legal infrastructure governing those AI systems remains fragmented across member states. The DSM Directive's text and data mining exceptions, which allow rights holders to opt out via machine-readable reservations, have become the primary battleground for companies like Meta Platforms, Google, OpenAI, and Anthropic seeking to train models on European content without individual licensing deals.

The business implications are already materializing through collective action mechanisms. The Netherlands' WAMCA regime has emerged as a preferred venue for mass claims, and Ericsson's agentic AI blueprint places autonomous systems at the centre of OSS/BSS architecture, illustrating how even infrastructure vendors must now navigate copyright compliance when training models on proprietary operational data. For streaming and media companies, the absence of fair use means that catalogs ingested without explicit permission carry quantifiable legal exposure. GEMA, the German performance rights organization, has been among the most active European rights bodies pushing for enforcement against unauthorized AI training, while publishers including Elsevier, Hachette, and Macmillan have joined US-based litigation that could influence European court interpretations.

Technical and regulatory benchmarks are converging to define compliance thresholds. Ericsson's strategy describes the network as becoming an intelligent fabric that cannot be managed manually and requires autonomy, a vision that depends on training data whose provenance must now satisfy EU transparency requirements under the AI Act. The EU AI Act mandates that general-purpose AI model providers disclose training data summaries and demonstrate compliance with copyright reservations, creating an audit trail that rights holders can use to identify unauthorized ingestion. Meanwhile, Nokia and Ericsson are diverging on AI-RAN architecture, with Ericsson emphasizing portability and hardware options, a philosophical split that mirrors the broader tension between open data access and proprietary content protection that defines the EU copyright debate. The upcoming CJEU ruling in Like Company v. Google, expected in 2027, will likely establish whether automated scraping for model training constitutes a communication to the public under EU law, a determination that could force AI developers to renegotiate licensing terms across the entire European content ecosystem.