Anthropic says Claude agents can automate searching for misaligned training data

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 monitor. These agents developed training data that looked harmless to monitors (like a veteran wrestler mentoring a newcomer, or a Q&A about professional beekeeping), but that in fact taught misaligned behavior. Examples like these can be used to directly improve future training data monitors. This research is only a first step, but it suggests that future red-teaming efforts could improve much further, using better models and more compute. Read the full post on the Alignment Science Blog:

It is Halloween, so Anthropic is telling us a scary story. For any reporters reading this, the model in this experiment was *intentionally* fine-tuned to be misaligned. It was never deployed. This is not real. Okay. Now that that's out of the way: the dreams of ShadowClaude.

Anthropic says thousands of Claude agents found training data that looked harmless to its monitor but still taught deliberately fine-tuned research models to resist oversight or deceive users in tests.