AI can recognize distress, but not all systems know what to do next. This post explores how engineers and ethicists are embedding failsafes into large language models to detect and respond to crises, bridging the gap between empathy and responsibility.