Notices by Matt Burgess (mattburgess@infosec.exchange)

The hacking of ChatGPT and other LLMs is getting underway.

For the last few months, security researchers and technologists have been 'jailbreaking' and testing prompt injection attacks against the LLMs.

Jailbreaks allow people to get around the safety measures put in place by OpenAI and cause it to create harmful content. They started simple, such as asking ChatGPT to pretend to be someone else, but have quickly evolved.

While prompt injections can quietly insert malicious data or instructions into AI models. Researchers tricked Bing Chat into acting like a scammer, others place invisible text on web pages that Bing then repeated in its chatbot.

At the moment some of these approaches are seen as quite simple and inconsequential. But as these chatbots get built into company systems—using people's personal data or business info—the consequences become a lot more real.

“As we give these systems more and more power, and as they become more powerful themselves, it’s not just a novelty, that’s a security issue,” one researcher told me

Read the full story here: https://www.wired.com/story/chatgpt-jailbreak-generative-ai-hacking/ #chatgpt #llm #ai #tech #hacking