OpenAI releases a free NCMEC-built CSAM image classifier and adds ChatGPT lockouts

The classifier targets the hardest category: novel and synthetic imagery that hash-matching systems such as PhotoDNA cannot catch. Distributing it at no cost lowers the barrier for smaller platforms that cannot afford in-house trust-and-safety ML.
The reported metrics matter for operations. Cutting false-positive reviews by 48% while keeping full recall directly reduces exposure for human moderators, one of the most damaging jobs in the industry.
On the product side, ChatGPT now steps up its response to repeated violative prompts. It moves from warning screens to support redirects to 24-hour lockouts. OpenAI is also publishing developer guidance.
The release lands the same day as California's lawsuit, and some observers read it as a reputational counterweight. On HN, the child-safety post drew 108 points and 241 comments, with debate over privacy and access controls for a classifier that is effectively a CSAM detector. Watch for which platforms adopt it and whether OpenAI publishes independent audit results.