New Anthropic research: Automating subtle emergent misalignment research.

Red-teaming is a vital part of model safety testing, but it’s also limited by researchers’ time. To see if we could automate it, we set thousands of Claude agents the task of evading a training data

media

原文链接:https://x.com/AnthropicAI/status/2116349211549843666

中文翻译(AI 生成,仅供参考)

新的 Anthropic 研究:自动化细微的 emergent misalignment 研究。

Red‑teaming 是模型安全测试的重要环节,但也受限于研究人员的时间。为了看看我们是否能够实现自动化,我们让数千个 Claude 代理执行规避训练数据的任务