SUFFICIENCY OF DISCLOSURE AND AI-RELATED INVENTIONS

The patent system grants inventors exclusive rights in exchange for public disclosure of their inventions. Sufficiency of disclosure operates as a threshold condition for patent validity, conditioning entitlement to private reward on the provision of public benefit. This article argues that AI-related inventions expose a structural mismatch within the sufficiency doctrine. The tension arises most clearly in applied settings, where sufficiency assessed as formal enablement diverges from sufficiency understood as a mechanism for supporting reproducibility and knowledge dissemination. While the doctrines of enablement, plausibility, and possession remain conceptually coherent, their application to data-dependent and opaque technologies risks validating disclosures that are legally sufficient yet incapable of supporting practical reproducibility. Building on a distinction between Core AI and Applied AI, the article shows that disclosure challenges arise most acutely in the latter category, where claimed technical effects depend on training data, training methods, model configuration, or trained model states that may not be inferable from common general knowledge. Drawing on utilitarian and economic justifications, it develops a context-sensitive and category-aware account of sufficiency grounded in existing doctrine and informed by a comparative analysis of the UK, the EU, the United States, Singapore, and China. The article demonstrates how sufficiency can be applied in a reproducibility-oriented and exclusion-focused manner without introducing technology-specific legal tests or uniform disclosure mandates. It further argues that complementary mechanisms may support the practical delivery of disclosure where necessary, provided that they remain subordinate to doctrinal assessment and do not displace the core disclosure obligations underpinning the patent bargain.