AI training raises serious copyright and IP issues for creators, with potential for substantial damages in U.S. lawsuits against AI firms as experts warn.

The core legal questions revolve around whether rights-holder permission is required or if textual data mining exceptions apply, and there is no uniform transparency standard across the EU, UK, and other jurisdictions.

France is pushing a presumption of use for training data, shifting the burden of proof to the AI developers, a model some officials say could be adopted EU-wide to streamline enforcement.

Concerns about sourcing content from pirated or illegal origins threaten fair-use arguments and complicate the construction of training datasets.

Memorization in AI models is studied, but the legal focus is on whether reproduction happened and from which sources, not simply whether data remains memorized.

Celebrity voices and likenesses are at the center of action, with figures like Taylor Swift and Matthew McConaughey pursuing voice trademarks and exploring personality rights, GDPR protections, and image rights as defenses.

EU text-mining exceptions exist but come with strict conditions—legitimate access, primarily non-profit research use—and may not cover all AI training phases, fueling debates about fair-use analogies.

The current legal landscape includes U.S. class actions and large settlements, such as Anthropic’s $1.5 billion deal, underscoring the risk of multibillion-dollar damages over training-data use.

EU regulatory directions may favor the French model that holds developers, not authors, accountable, potentially within an updated Intellectual Property Rights Enforcement Directive to harmonize rules across member states.