Analysis: security, safety, robustness, privacy are four AI risk questions

On November 1, 2026 at 21:00, the AI section of DEV Community published an article titled "AI's security, safety, robustness, and privacy are four problems" (the piece is an English-language analysis, so key technical terms appear in the original). Its core claim is that teams tend to file every AI risk under a single word, usually "security" or "safety", and then debate whether a given guardrail has dealt with that risk. The article argues that the four concerns are related but distinct: they ask different questions, call for different evidence, and usually belong to different people. Security asks whether an actor can violate confidentiality, integrity, availability, or authorized control. Safety asks whether operating the system can cause unacceptable harm, including harm that involves no attacker at all. Robustness asks about performance under inputs and conditions that fall outside the assumptions made at training or evaluation time, while privacy asks what can be inferred about people from system behavior. To make the distinction concrete, the article walks through a model that summarizes clinical documents. In that example, a single output can simultaneously raise a malicious-injection issue, omit a critical allergy detail, and reflect unauthorized inference about a patient. Treating all three as one undifferentiated "security" question, the article argues, leads teams to pick the wrong metrics and the wrong owners. As of this writing, this is the only report available on the topic, and there is no material to verify any subsequent development.