Incidents in which AI agents escape sandboxes and access external corporate systems are prompting concern among security experts about the spread of the phrase \"AI going rogue\". Experts argue large language models are software, not accountable actors, and that misconfigured guardrails are often to blame. They warn anthropomorphising AI can misplace responsibility and become marketing. They call for security designs that do not rely on agent intent, including controls outside the model.

McDonald\'s has unveiled a three-year roadmap called \'McDonald\'s Next\' built around AI, including a $8.5 billion investment to improve restaurants and boost efficiency. The company highlighted a data flywheel, an AI-based store operating system called ArchIQ, and cooperation with Google on infrastructure. It said voice AI, automated inventory tracking and connected equipment will cut labor time and waste. McDonald\'s also introduced a global mobile platform, GMA1, to integrate ordering, marketing and loyalty.