Subject: Critical Input Injection Error – Phantom Message Appeared Without User Action
Platform: ChatGPT (iOS App - iPad Pro 13”)
Date/Time: May 4, 2025, Approx 1:45AM EDT
Model in Use: GPT-4 / GPT-4.5 (active session – not manually switched)
Conversation Title: None. Self-Injected, Self-Initiated Prompt
Issue Description:
During a normal session, a full paragraph of text spontaneously appeared in the input history without being typed, dictated, or pasted by me. I did not speak, dictate, paste, or interact in any way. Voice and Transcription were not being used at the exact time of incident. There were no videos being played. The phrase repeated the same line multiple times:
“In this video, I will be showing you how to create a chatbot in ChatGPT, OpenAI, DALL·E, GPT-3, and GPT-4.”
This occurred approximately 10–15 times in a row, in one uninterrupted block of text. I did not trigger it by voice, clipboard, or macro. This was not a hallucinated output, but an unauthorized input that appeared inside the prompt box or as the user message itself, without any action on my part.
Implications:
This is a critical system error that violates session integrity. Possible causes include:
- Input queue corruption or overwrite inside the iOS app,
- Session bleed or memory leak across user accounts,
- Misrouted cached prompt injection from another user.
This raises serious concerns about data isolation, session privacy, and assistant reliability.
Requested Action:
- Escalate to engineering for forensic analysis.
- Check backend logs for any unusual input behavior tied to my account/device.
- Investigate any correlated prompt collisions or phantom inputs on the platform.
- Notify me of any findings or steps taken to isolate the bug.
This is a high-priority integrity and trust issue that deserves immediate triage.
Reported by: Dave Young
Device: iPad Pro 13” M1 Cpu
Model version: ChatGPT GPT-4/GPT-4.5
Browser/App: iOS ChatGPT App