长文本中语言模型会漏检危险行为，越长越容易出错。
Classifier Context Rot: Monitor Performance Degrades with Context Length
- 用提示词定期提醒，缓解长文本中的漏检问题
- 在800K无害内容后，危险行为漏检率提升2至30倍
- 适合关注AI安全监控的开发者与评估者
使用大语言模型监控代码代理的危险行为需处理超50万词元的对话记录，但现有基准测试大多不超过10万词元。本文发现，当前前沿模型在作为分类器时，对长文本中的危险行为识别能力显著下降。在需识别隐蔽危险行为的数据集上，Opus 4.6、GPT 5.4和Gemini 3.1在800K词元无害内容之后出现的危险行为，漏检率比独立出现时高出2至30倍。研究还表明，通过在长文本中周期性插入提醒提示词可部分缓解该问题，未来更好的后训练策略或能进一步改善。忽视长上下文性能退化的评估可能高估了监控系统的真实表现。
原文摘要 · Abstract (English)
Monitoring coding agents for dangerous behavior using language models requires classifying transcripts that often exceed 500K tokens, but prior agent monitoring benchmarks rarely contain transcripts longer than 100K tokens. We show that when used as classifiers, current frontier models fail to notice dangerous actions more often in longer transcripts. In particular, on a dataset that requires identifying when a coding agent takes a subtly dangerous action, Opus 4.6, GPT 5.4, and Gemini 3.1 miss these actions $2\times$ to $30\times$ more often when they occur after 800K tokens of benign activity than when they occur on their own. We also show that these weaknesses can be partially mitigated with prompting techniques such as periodic reminders throughout the transcript and may be mitigated further with better post-training. Monitor evaluations that do not consider long-context degradation are likely overestimating monitor performance.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。