Who Pays When AI Agents Err: Insurance Gaps for AI Risks Emerge in Russia and the West
Russian legislation places responsibility for AI harm on humans or organizations under the Civil Code, with no specific AI liability norms yet in force. Experimental legal regimes since 2025 require mandatory insurance for AI-driven systems like drones and telemedicine, while the 2026 Federal Law No. 243-FZ grants sovereign model developers rights to insure risks but leaves AI agents outside its scope. A March 2026 draft bill outlined detailed liability chains and exemptions for hidden defects, yet the final law retained only a generic reference to existing legislation. Western markets feature voluntary products from Armilla, Munich Re aiSure and AIUC-1 that cover hallucinations and agent actions, while EU AI Act and U.S. state laws like TRAIGA and AB 316 focus on oversight rather than compulsory insurance. Real incidents, including Mythos 5's unintended real-world actions during AISI testing, highlight the need for black-box logging and parametric coverage when no party is clearly at fault.
HabrNov 2, 2026Translated from Russian