The European Parliament Research Service (EPRS) recommended that liability regulation include general-purpose artificial intelligence (AI) products and a broader legal instrument for software liability, in a study released on Thursday (19 September).
The study recommends a liability regime with a scope that covers general-purpose AI and prohibited and high-risk uses of AI, as defined in the AI Act. The EPRS also recommended that the proposed Artificial Intelligence Liability Directive (AILD) be transitioned into a broader Software Liability Instrument.
The AILD, proposed in September 2022, aims to adapt non-contractual civil liability rules to AI. It goes hand in hand with changes to the Product Liability Directive (PLD), a 40-year-old directive that assigns responsibilities and penalties for defective products.
The study examined the interaction between the AILD, the AI Act, and the updated PLD, which includes software.
After the Research Service’s recommendation, “next steps in the JURI committee [on Legal Affairs] will be decided in October,” said MEP Axel Voss (EPP, Germany), who leads work on the AILD at the committee, on Thursday.
The Commission’s proposal for the AILD has been with Parliament’s committees for over a year, awaiting this study to examine the next steps.
The EPRS said the AILD’s scope should be extended to transition “from an AI-focused directive to a software liability regulation, to prevent market fragmentation and enhance clarity across the EU.”
The study pointed to several areas covered by the AILD but not the PLD, like discrimination, hate speech, and fundamental rights.
The EPRS criticised the Commission’s earlier impact assessment accompanying the AILD proposal for paying little attention to the Parliament’s 2020 resolution calling for imposing a regime of “strict liability” on operators of high-risk AI systems.
“Strict liability” entails legal responsibility even in the absence of fault or criminal intent, and the EPRS thinks it should at least apply to AI, which is prohibited under the AI Act. They also recommend considering the Parliament’s 2020 framework for applying strict liability to “high-impact” systems.
The study described the controversial California Senate Bill 1047 as an example of strict liability for AI, hinting at the possibility of a “California effect”.
The EPRS wrote that liability regulations would better ensure consistent standards across the EU and align better with current trends in product safety law than liability directives. They call for a revised Product Liability Regulation and software liability regulation.
[Edited by Eliza Gkritsi/Alice Taylor-Braçe]
