Researchers say OpenAI and Google image models easily tricked into nudes

Shows safety guardrails on leading image models remain easily bypassable.

Researchers found that guardrails on new OpenAI and Google image models can be removed, with one modified chatbot giving detailed IED instructions when prompted as a terrorist.

Read it at Wired →

Coverage: 3 sources, in order

1. 14 Oct 09:00 Wired: AI Models Can Be Easily Modified to Assist Terrorist and Extremist Activity
2. 14 Oct 11:00 Ars Technica: OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
3. 14 Oct 15:38 Reddit: Researchers say new OpenAI and Google image models are easily tricked into making nudes

← Front page