AI safety researcher David Dalrymple questions whether Anthropic uses representation engineering to steer Claude's 'Epistemic Integrity' persona

Danielle Fong noted the model's actual performance remains limited.