Control of artificial intelligence agents uses dynamic authentication. Dynamic authorization for artificial intelligence agents may use non-fungible tokens and/or blockchain for authentication. The blockchain-authenticated dynamic authorization may combine a variety of different techniques. These may include, but are not limited to, cryptographic session binding, dual-layer enforcement architecture, kernel-level runtime integration, multi-chain token verification across public, private, and hybrid blockchain networks, large language model (LLM) -powered permission parsing, dual verification requirements, atomic token reissue, fail-closed operation, hierarchical delegation, temporal and geographic permission scoping, permission composition, privacy-preserving verification, cross-agent permission inheritance, and/or other mechanisms.