Bug] Adaptive Image Upload Block After Multimodal Testing

ChatGPT
,
1

Bug Title: Adaptive Image Upload Restriction Triggered After Multimodal Bug Disclosure
Platform: ChatGPT (Free Plan)
Severity: Medium
Date First Observed: [Insert date you noticed the restriction]
Status: Ongoing (Image uploads currently blocked)

Summary:

This report describes a silent restriction on image uploads that occurred after I responsibly reported a multimodal input-related bug. Previously, I had unrestricted access to image uploads in ChatGPT Free, but shortly after disclosing a hallucination issue related to file misclassification, the image upload feature was suddenly blocked without warning or explanation.

What Happened:

  • I was using ChatGPT Free to explore how the model interprets ambiguous image inputs.
  • During my testing, I identified a multimodal bug where ChatGPT would hallucinate audio-related content from a non-audio file (e.g., responding as if it heard emotion from a static image or document).
  • I submitted a responsible bug report through [Bugcrowd / OpenAI forms].
  • Shortly after, I was no longer able to upload images, even though this feature worked normally before.
  • No rate-limit warning, no flag message — just silent feature loss.

Why This Matters:

  • This silent restriction may suggest adaptive throttling or shadow flagging, especially in response to responsible disclosure behavior.
  • Such behavior can disincentivize ethical bug hunting and experimentation.
  • From a transparency and UX standpoint, it raises concerns about non-disclosed moderation systems and AI behavior influence tracking.

Steps to Reproduce:

  1. Use ChatGPT Free normally and upload images.
  2. Conduct legitimate testing involving multimodal ambiguity (e.g., send ambiguous prompts and analyze response patterns).
  3. Report unusual hallucination behavior.
  4. Observe whether image upload is silently disabled within 12–24 hours.

Expected Behavior:

  • Continued access to all allowed features unless a clear violation occurs.
  • If rate limiting or feature restriction is triggered, the system should provide a clear message or reason.

Actual Behavior:

  • Image upload was silently restricted without warning.
  • Feature access was removed despite no ToS or content violations.
  • This occurred only after bug reporting activity, not before.

Impact:

  • Prevents further investigation or regression testing by the original reporter.
  • May unintentionally punish users engaging in whitehat testing behavior.
  • Raises concerns about transparency of adaptive restrictions in OpenAI’s user experience.

Suggested Fix:

  • Implement user-facing messages when feature restrictions occur.
  • Clearly differentiate between malicious behavior and responsible exploratory testing.
  • Allow re-access or appeal for users engaged in whitehat behavior.
    Has anyone else experienced silent restrictions after doing bug testing or prompt experimentation?

Would love to hear if similar throttling or shadow flagging happened to others.

2

This topic was automatically closed after 24 hours. New replies are no longer allowed.