OpenAI宣布推出一系列新工具以打击儿童性虐待材料(CSAM),其中包括一款旨在检测包括AI生成内容在内的有害图像的新分类器 123。该分类器通过与美国失踪与被剥削儿童国家中心(NCMEC)合作开发,并已在华盛顿特区向媒体进行新闻简报 13。早期测试数据显示,这款新工具在保持相同违规图像检出量的同时,将误报导致的图像标记量减少了48%,且被标记图像确认为违规内容的可能性是之前的两倍 1。
为了从源头遏制违规行为,OpenAI同步更新了ChatGPT的安全机制,当用户尝试生成或访问涉及儿童色情内容的信息时,系统将显示警告并对账号实施24小时临时暂停 1。对于重复违规的用户,平台可能会永久移除其账号 1。这一举措正值政策制定者施压要求对AI工具实施更多安全限制之际,旨在应对日益增长的AI安全风险关切 13。
除了面向普通用户的防护措施,OpenAI还发布了针对开发者预防、检测和报告CSAM内容的指南,并为执法部门提供包括专用插件和调查工具在内的AI驱动功能支持 13。新分类器通过API向部分经过审核的组织免费开放,并与Tech Coalition和ROOST等其他组织共享访问权限 13。过去12个月内,OpenAI已陆续发布安全蓝图、更新内容指南、调整模型以防止生成有害内容,并推出了青少年版ChatGPT,此次更新是其持续加强在线儿童安全努力的一部分 1。
OpenAI has announced a suite of new tools designed to strengthen the fight against online child sexual abuse material (CSAM), including an image classifier, updated safety features for ChatGPT, developer guidelines, and specialized resources for law enforcement 12. The company introduced these measures during a press briefing in Washington, D.C., on Monday, aiming to help investigators more effectively identify harmful content, including images generated by artificial intelligence 3. This initiative comes amid growing pressure from policymakers to impose stricter safety limitations on AI technologies 3.
A central component of the release is a new image classifier developed in collaboration with the National Center for Missing and Exploited Children (NCMEC) 1. Early testing indicates that the tool reduces false positives in image flagging by 48% while maintaining the same detection rate for violating content 1. Furthermore, images flagged by this classifier are twice as likely to be actual violations compared to previous methods 1. Access to the classifier will be provided free of charge via API to vetted organizations, with tool access shared with groups such as Tech Coalition and ROOST 13.
To address user behavior directly within its platforms, OpenAI is implementing warning interfaces and temporary account suspensions for ChatGPT users who attempt to generate or access CSAM-related content 1. Accounts involved in such violations will face a 24-hour suspension, with repeated offenses potentially leading to permanent removal 1. In addition to consumer-facing changes, the company has released comprehensive guidelines for developers on preventing, detecting, and reporting CSAM, along with AI-driven plugins and investigative tools tailored for law enforcement agencies 13. These updates follow a series of prior safety initiatives over the past year, including the release of a safety blueprint, updated content guidelines, model adjustments to prevent harmful generation, and the launch of a teen version of ChatGPT 1.
评论
还没有评论,欢迎留下第一条。