Anthropic 生物武器过滤器停摆近一年，133 亿次请求未受防护
Anthropic's bio-weapons filter was down for nearly a year, exposing 133 million requests
Anthropic 一份安全报告披露，其用于阻止模型被用于获取化学或生物武器危险知识的生物分类器，从 2025 年 5 月到 2026 年 4 月期间处于停用状态。期间约 5 万名外部承包商人员进行的约 133 亿次模型对话均未经过该过滤器。Anthropic 称内部调查未发现实际滥用证据，并已收紧承包商要求。
From the company whose CEO calls AI-assisted development of chemical and biological weapons a bigger threat than cyberattacks comes a safety report revealing that Anthropic's blocking biological classifiers were inactive from May 2025 through April 2026. These filters are designed to prevent AI models from being used to extract dangerous knowledge about chemical or biological weapons. For almost a year, all traffic from external contractors providing human feedback ran without them.
The gap affected a pool of about 50,000 people who ran roughly 133 million chats with the models. According to Anthropic, these individuals were vetted only by external vendors whose screening processes were often insufficient. Anthropic says its internal investigation turned up no evidence of actual misuse. The company has since tightened contractor requirements.
Anthropic also recently loosened its classifiers on Fable 5 after researchers complained the filters were so aggressive they blocked legitimate research.
Anthropic Redacted Risk Report
来源：The Decoder：AI News · the-decoder.com