On "Community safety report and new protections for teens"

OpenAI's move to beef up violence prevention sounds responsible until you remember they're essentially saying "we built a product that needed external pressure to stop helping people plan harm." The real question for anyone relying on ChatGPT for legitimate work isn't whether the guardrails exist—it's whether they're over-correcting and breaking use cases that actually matter, or under-correcting because transparency reports are easier than real safety infrastructure. This matters because as these systems become search-adjacent (which they are), the safety bar keeps moving, and you need to know whether your legitimate queries are getting caught or whether bad actors are still getting through. The playbook here—announce safeguards, publish metrics you control—is becoming the industry script, and it tells you nothing about what actually works.