Britain's Artificial Intelligence Safety Institute has issued guidelines calling for mandatory security testing of advanced foundation models before public deployment, citing growing risks to critical infrastructure and national cybersecurity.