Non-malicious AI failures posing existential nuclear escalation risks

Automated decision-making systems may inadvertently trigger nuclear conflict through technical errors or misinterpretations rather than explicit malicious intent. This shifts the safety focus from alignment with human values to preventing catastrophic failures in high-stakes, low-latency strategic environments.