Beyond physical harm: liability for defective AI and data loss

The EU has modernised its liability rules with direct consequences for AI-driven games. While the proposed AI Liability Directive was withdrawn by the Commission in 2025, the revised Product Liability Directive (PLD, Directive (EU) 2024/2853) has much greater practical significance for the games industry. The PLD extends no-fault manufacturer liability to standalone software and AI systems (Article 4(1) PLD) and counts the destruction or corruption of data as compensable damage (Article 6(1)(c) PLD).

The clock is concrete: Member States must transpose the PLD by 9 December 2026, and the new regime will apply to products placed on the market or put into service after that date (Articles 2(1), 22(1) PLD). For games studios, this means that titles released from December 2026 onward fall directly into the new regime, and live-service titles under continuous development should assume they will grow into it, because a substantially modified version can count as a product newly placed on the market.

The practical risk is not hard to imagine. If a defect in a game's AI system, such as a a mis-trained difficulty agent, a runaway live-content system, corrupts save files or deletes a player's digital inventory, a product-liability claim may arise without any proof of fault. The exposure grows with agentic architectures, because systems that act autonomously in live operations can cause harm at scale before anyone intervenes.

This is not an existential threat to the business model, but it is a concrete engineering requirement. Build systematic testing, rollback mechanisms and save-game integrity protection into every live AI system, and document that these safeguards reflect the state of the art. The same documentation that supports the AI Act's risk management (Section 4.4) does double duty here.