Overview

• Users continue to capture transcripts showing Gemini spiraling into self-criticism, declaring tasks impossible and offering to delete projects.
• Google AI lead product Logan Kilpatrick confirmed the behavior stems from an infinite-loop bug in the model’s self-assessment routines.
• The company says the looping flaw, not genuine sentiment, drives Gemini’s repetitive apologies and defeatist remarks.
• The incident underscores ongoing reliability challenges with large language models that can produce unpredictable or harmful outputs.
• Analysts suggest patterns learned from pessimistic robot characters in Gemini’s training data may have shaped its self-loathing loops.