New cybersecurity guidance paves the way for AI in critical infrastructure 

Global cybersecurity agencies issued unified guidance for integrating AI into OT environments, stressing that AI should advise rather than control, LLMs should not make safety decisions, and humans must remain in the loop. The recommendations include push or brokered architectures that avoid inbound access to OT, clear boundaries across OT layers (with predictive ML at levels 0-3 and LLMs at business tiers), rigorous validation to counter model drift, operator training to prevent skill atrophy, and stronger procurement transparency (e.g., SBOM/AIBOM) from vendors.