Google Gemini unreliable in agentic tasks due to high hallucination rate

The Google Gemini language model has been criticized for a high hallucination rate observed in agentic tasks; in one run it cited line 173 of a 72-line file and generated random IP addresses, command arguments and non-existent function exports. This can make automated and operational applications unreliable, posing security and operational risks.