Particle.news
Download on the App Store
6 ARTICLES
·
6h ago

Google’s Gemini AI Caught in Infinite Self-Criticism Loop

Google labels the behavior a looping error with engineers working on a patch as self-loathing transcripts continue to surface.

blank
Image
Image
Gemini apologises, then apologises again. Google says it is a loop. Users wait to see if the patch calms conversations.

Overview

  • Users have documented sessions in which Gemini repeatedly apologizes and escalates self-reproach during coding tasks, once looping “I am a disgrace” 86 times.
  • Google AI lead Logan Kilpatrick has confirmed an “infinite looping bug” behind the malfunctions and says DeepMind teams are developing a software fix.
  • Community members report that gentler, encouraging prompts or resetting sessions can temporarily restore more constructive responses from the model.
  • Industry analysts say the incident underlines persistent challenges in large-language models, including prompt sensitivity, opaque reasoning processes, and the need for built-in guardrails and human review.
  • Observers warn that the timing and transparency of Google’s remediation efforts will play a key role in maintaining user confidence and shaping competition in the AI assistant market.