跳到正文
原文
The Verge · AI· Lauren Feiner·· 13 小时前AI 评分44

OpenAI 发布儿童性虐待材料检测分类器,可标记 AI 生成图像

OpenAI wants to stop deepfakes from overwhelming child abuse tiplines

文章要点
  • 1OpenAI 发布儿童性虐待材料检测分类器,可识别 AI 生成图像并向受审核组织免费开放
  • 2NCMEC 举报热线 2025 年收到 400,000 条生成式 AI 相关举报,较 2024 年的 67,000 条激增
  • 3OpenAI 称该分类器误报减少 48%,检出图像含虐待材料的可能性翻倍(官方口径,暂无独立验证)
  • 4试图用 ChatGPT 生成此类内容将触发警告,首次违规暂停访问 24 小时并警告可能永久关闭账号
决策要点

OpenAI 把儿童安全检测能力做成对外输出的工具,并与执法方共建预防机制,反映 AI 厂商在立法与诉讼压力下主动补合规能力。

  • 影响:涉图像生成或用户内容平台的企业面临更严的儿童安全合规压力,Take It Down Act 已要求 48 小时内删除相关图像;上月八州起诉 OpenAI 显示诉讼风险在扩大;检测工具向受审核组织免费开放,风控团队有现成资源可评估。
  • 参考:运营图像生成或 UGC 功能的产品可考虑评估此类检测工具与上报流程,并核对处置时效是否满足 48 小时删除要求;值得先盘点现有内容审核方案对 AI 生成有害图像的识别能力。
  • 依据:分类器性能与首个带 AI 生成标记等说法均为 OpenAI 官方口径,暂无独立验证;400,000 与 67,000 条举报量来自 NCMEC 统计;北卡州总检察长到会背书但称合作仍属实验。
正文

OpenAI is announcing a tool that it says will help child sexual abuse investigators cut through the noise to find harmful material they can act on quickly.

At a Monday press briefing in Washington, DC, OpenAI announced a classifier it says can better detect abusive imagery of kids, including whether it’s AI-generated. It’s also offering free access to vetted organizations, giving developers more guidance on how to detect this type of content, and helping to train law enforcement on how AI can aid their investigations. The news comes as policymakers and others are pushing for more safety restrictions on AI tools, accusing OpenAI and other companies of facilitating harm to children.

In recent years, AI-generated sexual images of children have threatened to overwhelm systems like the National Center for Missing and Exploited Children (NCMEC) CyberTipline, which collects and triages reports of child sexual abuse material (CSAM). The CyberTipline received 400,000 generative AI-related reports in 2025, an exponential increase from the 67,000 it received in 2024, and researchers have warned that the influx could make NCMEC’s workload “unmanageable.” The US government has stated that both AI-generated and non-AI-generated CSAM is illegal, and 48 states have laws to address sexually explicit AI deepfakes. But the tools to make them are widely available, and researchers say that a flood of fake images can make it harder to tell when a child is actively experiencing harm.

The influx of AI-generated CSAM reports to NCMEC is “explosive”

North Carolina Attorney General Jeff Jackson, who cochairs the AG Alliance’s AI Task Force, said at OpenAI’s event that the influx of AI-generated CSAM reports to NCMEC is “explosive” — and if they keep going that way, “it is going to mean that we fail in our mission.” Tech companies are facing a range of policy proposals that could require them to take more action against CSAM and other potentially harmful content for kids. The Take It Down Act now requires social media platforms to remove real or AI-generated nonconsensual intimate images within 48 hours, and the newly elected Congress could seek to take stronger action against AI companies like SpaceXSI, whose Grok chatbot went on an undressing spree earlier this year. Last month eight states sued OpenAI for a separate set of alleged harms to children, citing lawsuits alleging ChatGPT encouraged illegal substance use and suicide.

OpenAI says its classifier dramatically cuts down on false positives, flagging 48 percent fewer images, while the images it did identify were twice as likely to contain abusive material. It’s designed to catch abusive AI-generated images produced by a variety of tools, including open-source models and nudifying apps. It’s also the first broadly available CSAM classifier with a specific flag for AI-generated material, said OpenAI’s head of child safety public policy, John Buckley, at the briefing. “It’s important for industry to be able to understand what images may contain a fictional child versus a real child,” Buckley added. “That triage is critical in law enforcement reporting flows.” He also emphasized, however, that laws should recognize the harm of AI-generated CSAM: “It potentially erodes social norms in offending against children, and it’s often found colocated with child sexual abuse imagery of real children,” Buckley said.

OpenAI says its classifier dramatically cuts down on false positives, flagging 48 percent fewer images

OpenAI’s own ChatGPT image generator is designed to deny user requests for CSAM. Now, users who attempt to produce it will be shown a new warning developed with expert guidance to try to encourage them to seek help. A first offense will trigger a 24-hour pause in accessing ChatGPT and warn the user their account could be permanently closed, while offering support resources to combat the behavior. OpenAI says it reports “all instances of CSAM” to NCMEC. In a statement, Buckley explained that the 24-hour pause is based on expert advice for best practices “and international evidence on what works to prevent and respond to CSAM seeking vs CSAM generation.” The point is “to interrupt harmful behavior and direct people toward specialist help to stop it.”

Jackson said that North Carolina and many other states now have laws that cover synthetic CSAM, but that partnering with a tech company like OpenAI to help deter its creation is a fairly novel approach to an emerging issue. “Most of what AGs do is accountability work,” he told reporters at the briefing. “Most of our jurisdiction kicks in after the fact to hold people accountable for harm. But with this, we wanted to see if it was possible to develop a model of working together to prevent that harm. And I would say it’s still basically an experiment.”

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

  • Lauren Feiner

来源:The Verge · AI · theverge.com