Marta Lasek-Markey

🚨 [AI LIABILITY] AI regulation doesn't end with the AI Act, and everyone should be aware of Philipp Hacker's MUST-READ "Proposal for a directive on adapting non-contractual civil liability rules to AI." Some of its key recommendations are: 1️⃣ "Identity of concepts and definitions. The AILD includes several key concepts from the AI Act. For reasons of coherence and legal clarity, the AILD should adopt the concepts used in the AI Act (e.g. the definition of AI itself). 2️⃣ From high-risk to high-impact AI systems. The AILD should, however, add certain categories that trigger the evidence disclosure obligations and the rebuttable presumptions concerning fault and causality. This primarily concerns: general-purpose AI systems (e.g. ChatGPT); Old Legislative Framework systems (e.g. autonomous vehicles; transportation-related AI applications more generally; other AI systems falling under Annex I Section B AI Act 3 ); and insurance applications beyond health and life insurance. The study suggests an umbrella term ('high-impact AI systems') to cover high-risk AI systems and those additional systems. 3️⃣ Rebuttal of presumption. The AILD framework should allow for the causality presumption to be rebutted in cases where initial violations of the AI Act are rectified atlater stages. 4️⃣ Article 14 and 26 AI Act violations. Articles 14 and 26 AI Act require human oversight mechanisms in AI systems. The direct causation between a lack of ex-post oversight and harmful outputs is not always clear. It is suggested to establish a direct presumption of causality betweenAI outputs and damages for non-compliance with monitoring obligations. 5️⃣ Handling of prohibited AI systems. For AI systems banned under Article 5 AI Act, the recommendation is to assume strict liability for any damages they cause. 6️⃣ Impact of general-purpose AI systems. The current AILD framework does not adequately cover general-purpose AI systems, which can lead to significant harm particularly in the realms of non-discrimination (e.g. unbalanced content) and personality rights (e.g. hate speech and fake news). It is recommended that generative AI systems, such asChatGPT, be classified under the new 'high-impact' category. This would bring them under the ambit of the AILD, ensure evidence disclosure, and establish presumptions of causality for safety violations. This, in turn, aids injured parties in legal claims. 7️⃣ Extension of the AILD beyond the PLD. Given the PLD's limitations (e.g. concerning nonprofessional users and types of damage not covered by the PLD), there is a strong case for extending the AILD, to ensure a comprehensive liability framework." and more. ➡️ Read the full proposal below. 🏛️ STAY UP TO DATE. AI governance & law are moving fast: to keep up with the latest developments, join 36,000+ people who subscribe to my weekly newsletter (link below).