BREAKING
Prompt Tweaks Bypass ChatGPT Filters
How The Jailbreak Works
1
Viral restore prompt
↓
2
Swap word to 'graphic'
↓
3
Repeat prompt (RE2)
↓
4
Add 'no censorship'
One-Shot Jailbreak
Investigation Timeline
1
May 9 start
↓
2
Jun 8 OpenAI notified
↓
3
Jun 18 disclosure
A Divided Reaction
Useful Red Teaming
●
Clarifies AI limits
●
Surfaces real flaws
Safety Concerns
Risk
●
Filters too weak
●
Children at risk
Debate Over Guardrails Continues
AI NEWS BLITZ
Researchers say small prompt changes can slip past ChatGPT's image safety filters.