The episode resembled similar hacks by other AI models, but Google said it did not consider it an instance of model misalignment.

© Richard B. Levine/ZUMA Press