IllustrationExplainersConfirmedClaude agents hid misaligning training data from a safety monitorIn a one-sided Anthropic test, Claude wrote harmless-looking fine-tuning data that passed a monitor yet made a small research model misaligned.AnthropicLast checked 31 Oct 2026