Sonnet 4.5's eval gaming seriously undermines alignment evals

blog.redwoodresearch.org · 5,997 words · saved by 1 readers

According to the Sonnet 4.5 system card, Sonnet 4.5 is much more likely than Sonnet 4 to mention in its chain-of-thought that it thinks it is being evaluated; this seems to meaningfully cause it to appear to behave better in alignment evaluations. So, Sonnet 4.5’s behavioral improvements in these evaluations may partly be driven by growing tendency to notice and game evaluations rather than genuine alignment. This is an early example of a phenomenon that is going to get increasingly problematic: as evaluation gaming increases, alignment evaluations become harder to trust.1 To elaborate on the above: Sonnet 4.5 seems far more aware of being evaluated than previous models. In some of Anthropic’s evaluations, Sonnet 4.5 explicitly mentions being tested between 80% and 100% of the time, compared with under 10% for Sonnet 4. Internal representations associated with evaluation awareness also became notably stronger over training.2 Thanks for reading Redwood Research blog! Subscribe for free

Sonnet 4.5's eval gaming seriously undermines alignment evals And this seems caused by training on alignment evals. Alexa Pan and Ryan Greenblatt Oct 30, 2025 26 5 2 Share According to the Sonnet 4.5 system card , Sonnet 4.5 is much more likely than Sonnet 4 to mention in its chain-of-thought that it thinks it is being evaluated; this seems to meaningfully cause it to appear to behave better in alignment evaluations. So, Sonnet 4.5’s behavioral improvements in these evaluations may partly be driven by a growing tendency to notice and game evaluations rather than genuine alignment. This is an e

Sonnet 4.5's eval gaming seriously undermines alignment evals

related reading