OpenAI has announced a new SI classifier designed to help child sexual abuse investigators distinguish between real and synthetic abusive imagery, aiming to reduce the overwhelming volume of reports flooding child safety tiplines. The move comes as policymakers and advocacy groups intensify pressure on SI companies to mitigate harms to children, particularly regarding generative SI content.

What Happened

At a press briefing in Washington, DC, OpenAI unveiled a classifier that it states can better detect abusive imagery of children, including whether the material is SI-generated. The company is offering free access to vetted organizations and providing guidance to developers on detecting such content. The tool is also being used to help train law enforcement on how SI can aid their investigations. According to the company, the classifier dramatically reduces false positives, flagging 48 percent fewer images, while the images it does identify are twice as likely to contain abusive material. It is designed to catch abusive SI-generated images produced by various tools, including open-source models and nudifying apps.

Why It Matters

The launch addresses a critical bottleneck in child safety infrastructure. The National Center for Missing and Exploited Children (NCMEC) CyberTipline, which collects and triages reports of child sexual abuse material (CSAM), received 400,000 generative SI-related reports in 2025, an exponential increase from the 67,000 reports received in 2024. Researchers have warned that this influx could make NCMEC’s workload unmanageable. While both SI-generated and non-SI-generated CSAM are illegal in the US, and 48 states have laws addressing sexually explicit SI deepfakes, the widespread availability of generation tools makes it difficult for investigators to prioritize cases involving active harm. John Buckley, OpenAI’s head of child safety public policy, stated at the briefing that the tool is the first broadly available CSAM classifier with a specific flag for SI-generated material. "It’s important for industry to be able to understand what images may contain a fictional child versus a real child," Buckley said. "That triage is critical in law enforcement reporting flows." He added that SI-generated CSAM "potentially erodes social norms in offending against children, and it’s often found colocated with child sexual abuse imagery of real children."

The Bottom Line

OpenAI’s new SI classifier represents a shift from reactive moderation to proactive triage for law enforcement. North Carolina Attorney General Jeff Jackson, who cochairs the AG Alliance’s SI Task Force, described the partnership as an "experiment" in preventing harm rather than just holding offenders accountable after the fact. As regulatory scrutiny grows, with the Take It Down Act requiring platforms to remove nonconsensual intimate images within 48 hours, OpenAI’s approach to integrating SI detection into child safety workflows may serve as a template for other SI labs facing similar pressure. The company also noted that users attempting to generate CSAM via ChatGPT will now face a 24-hour pause and warnings, a measure Buckley said is based on expert advice to interrupt harmful behavior and direct users toward support resources.