BREAKING
ChatGPT Image Tool Flagged by Mindgard
How the Filter Was Bypassed
1
Restore-photo prompt
↓
2
No image attached
↓
3
Swap words, repeat
↓
4
Harmful images
Accidental Prompt, Not an Attack
OpenAI vs Mindgard
OpenAI
Response
●
Blames no-image prompts
●
Encourages image uploads
●
Added new safeguards
Mindgard
Counter
●
Still works with tweaks
●
Calls it systemic
●
Filter not adequate
Probe May 9, Findings June 18
Safety Guardrails Questioned Again
AI NEWS BLITZ
Security firm Mindgard says ChatGPT made violent images from a simple prompt.