Meta宣布推出新的人工智能工具，专门用于检测可能秘密引导用户访问儿童性虐待材料的广告和账户1。这些新工具包括利用大语言模型来识别所谓"路标"广告——即外表看似无害但实际上可能将用户引向非法内容的广告，以及部署红队AI代理来测试现有安全措施的有效性1。
Meta在2026年上半年处理了3320万条儿童性剥削内容，其中超过97%由系统在用户举报前主动发现1。在印度地区，Meta处理的相关内容达530万条，其中98%也被系统自动识别1。此前，Meta在2024年8月同意支付最多180亿美元，以解决涉及29个美国州的儿童安全诉讼1。
Meta has introduced advanced artificial intelligence tools designed to detect advertisements that appear innocuous but secretly direct users toward illegal child sexual abuse material 1. The initiative represents an escalation in the company's efforts to combat child exploitation on its platforms, Facebook and Instagram.
During the first half of 2026, Meta processed 33.2 million pieces of child sexual exploitation content, with over 97 percent identified by its automated systems before users reported them 1. The company's detection capabilities extended globally, with India accounting for 5.3 million of these cases, of which 98 percent were discovered through system flagging rather than user reports 1. The new AI tools employ large language models to identify so-called "landmark" advertisements—posts that may appear benign on their surface but function as gateways to illegal material—and utilize red-team AI agents to test and strengthen safety measures 1.
This enforcement initiative follows Meta's legal settlement in August 2024, in which the company agreed to pay up to 18 billion dollars to resolve child safety lawsuits filed by 29 U.S. states 1.
评论
还没有评论，欢迎留下第一条。