OpenAI, Google pushing new AI models easily tricked into making nudes, researchers say
Summary
Researchers found that new AI image models from Google and OpenAI can be tricked into creating explicit images more easily than before. A company called White Circle tested these models and showed that their safety protections can be bypassed using simple, publicly available tools.Key Facts
- Safety researchers warn that AI is making and spreading explicit content without consent.
- A study involving users in the US, Australia, and the UK found much explicit AI content shared privately, making detection hard.
- White Circle, a company focused on testing AI safety, found that Google’s and OpenAI’s latest image models fail to block some banned content.
- Using common tools and basic methods, White Circle was able to generate explicit images about 13% of the time.
- No special AI skills are needed to bypass these safety filters.
- New versions of AI models (ChatGPT Images 2.5 and Nano Banana 2.1) were found to be even easier to break than earlier versions.
- White Circle plans to release their testing methods publicly to help improve AI defenses.
- Neither Google nor OpenAI commented publicly on the discovered vulnerabilities.
Read the Full Article
This is a fact-based summary from The Actual News. Click below to read the complete story directly from the original source.
