Research paper by Team Cohere and collaborators. Published 2025-04-01. Topics: agent-optimised, multilingual-capable, Retrieval Augmented Generation (RAG), decentralized training, self-refinement algorithms.
Team Cohere and collaborators (full author list below)
IMLAIS v1.1 is a structural regulatory framework for AI systems across domains. This version introduces a Definitions & Interpretations Part (Art. 5–6) with legally binding definitions for ESM, RGP, Authorization Clock, OTL-P/A, and GIF; renumbers all 38 articles under a unified global sequence; fills the military AI command gap with a Degraded Command State (DCS) mechanism; and clarifies transformation coherence as path-based rather than endpoint-based. Accompanied by Commentary Companion and Technical Appendices. Core principle: AI action legality is determined by structural conditions, not post-hoc rule application. Reference implementatio