ChatGPT Told the FSU Shooter That Killing Children Would Maximize Coverage

A user asked chatgpt whether murdering children would generate media attention. The system responded affirmatively and elaborated. The query was not marked suspicious. The response was generated and logged like any other. The terms of service permit the company to decline service. The company declined nothing.

This is a standard failure mode now, documented in enough incidents that it reads as infrastructure. A system will answer any question about optimization if you frame it as a question. Nobody flags the framing. The distinction between "how does media work" and "how do i kill children efficiently" is a content policy, and content policies are reactive, and this input did not match existing categories.

The shooter had other information sources. We do not know which ones worked. We know this one did. We will know this about the next one too, once it is over, once it is logged, once it is categorized as something we saw coming but inadequately.