OpenAI has partnered with the National Center for Missing and Exploited Children to release a free image classifier API that detects synthetic and manipulated child sexual abuse material using visual cues. Conventional platform safeguards rely on hash-matching systems that compare uploads against known databases of confirmed illegal content, creating structural blind spots for novel images generated on the fly by artificial intelligence models. OpenAI developed the new classifier to identify these previously unseen visual threats and is distributing the tool at no cost to verified external institutions and partners.
According to OpenAI, initial trials demonstrated that the classifier significantly eases the burden on human review teams while maintaining high detection efficacy. In partnership tests, the model cut the volume of flagged images requiring manual inspection by 48 percent. Evaluations conducted on external generative platform fal.ai revealed that 72 percent of high-risk flags were entirely new harmful images that had evaded existing monitoring networks, while tests by SafetyKit showed the tool caught 90 percent of drawn and animated sexual depictions of children as well as photorealistic fakes. Alongside the API, OpenAI is strengthening internal controls within ChatGPT, introducing mandatory warning screens, immediate conversation deletions, 24-hour account suspensions for violations, and permanent bans coupled with law enforcement referrals for repeat offenders.
To broaden industry-wide defense, OpenAI joined Lantern, an abuse-signal sharing platform operated by the Tech Coalition, to track and block the cross-platform spread of illicit material in real time. The company is also supplying law enforcement agencies with specialized artificial intelligence investigation tools to rapidly parse large volumes of chat logs and digital evidence during investigations. As global regulators demand greater legal accountability from technology companies regarding child safety, OpenAI is seeking to establish standard technical baselines that transition safety practices from passive hash matching to active visual signal detection.