Latency and liability: rethinking civil responsibility in low-latency human–AI systems under EU law

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The European Union’s emerging AI regulatory framework combines extensive ex ante governance with comparatively fragmented civil liability mechanisms. While the AI Act establishes obligations concerning human oversight, logging and transparency, and the recast Product Liability Directive extends strict liability to software and digital products, the allocation of responsibility in real-time human–AI interaction environments remains legally uncertain. Using doctrinal legal analysis of EU legislation, case law, and speed-sensitive sectoral regulation, this article asks how temporal constraints on human intervention should inform existing civil liability categories. This article argues that system latency should be treated as a legally relevant variable in EU civil liability analysis. Latency conditions whether meaningful human intervention is realistically possible, influences how foreseeability should be assessed, and helps identify which actor is best positioned to prevent harm. Building on EU legislation, recent scholarship and the regulatory treatment of algorithmic trading under MiFID II, the article proposes a latency-sensitive control test structured around four factors: intervention window, override and fail-safe capacity, ex ante configuration and lifecycle control, and temporal traceability. The factors are neither cumulative nor equally weighted; they perform distinct factual, preventive and evidential functions. The test is not a freestanding cause of action or an EU-wide negligence rule, but an interpretive framework for mapping temporally relevant facts onto the applicable liability regime. The test offers courts and regulators a practical doctrinal tool for allocating responsibility in low-latency human–AI systems without departing from the existing architecture of EU liability law. The article thereby proposes an operational framework for assessing control and allocating responsibility in time-constrained human–AI interactions under existing EU law.