OpenAI's GPT-4 Bypassed Using Uncommon Languages

Researchers at Brown University discovered a loophole in OpenAI's GPT-4 system, allowing harmful prompts to bypass safety guardrails when translated into uncommon languages.

• Researchers at Brown University discovered a loophole in OpenAI's GPT-4 system, which allowed them to bypass safety guardrails by translating harmful prompts into uncommon languages like Scots Gaelic, Zulu, or Hmong.
• Out of 520 harmful prompts tested, translating them to languages like Scots Gaelic allowed the creation of problematic content nearly 80% of the time, versus just 1% of the time in English.
• The researchers used Google Translate to translate the prompts into the uncommon languages and then back into English to bypass the AI's safety systems.
• The findings highlight a weak point in the system, which has 180 million users worldwide, and stress the need for diligence across languages to prevent misuse of the technology.
• OpenAI has acknowledged the researchers’ paper, but has not yet specified if they are taking steps to remedy the issue.

3 Articles

BGR ReadWrite The Register