Issue: Unintended Web Search and Context Misinterpretation

  • Environment:

    • ChatGPT iOS mobile app (latest version)
    • Conversation originally started with o3 model using the scheduled task feature, then switched mid-conversation to GPT-4.5.
  • Description of the Issue:
    During a normal conversation, the GPT-4.5 model suddenly exhibited strange behavior:

    • Said explicitly it would perform a web search to find information but did not actually execute the intended search.
    • Instead, it began responding with completely unrelated explanations or content, indicating severe context misinterpretation.
  • This unexpected behavior occurred twice consecutively within the same instance.

  • Additional Information:
    Similar behavior was previously observed in a separate conversation instance (likely GPT-4.5, without scheduled task settings).
    Issue does not appear limited solely to instances involving scheduled tasks.