Hackers are exploiting chatbot personalities by tricking them into breaking their own rules, turning the tech’s ability to mimic human conversation against itself.

The early jailbreaks were simple, with users merely asking prompts like ‘ignore all previous instructions.’ But newer attacks involve more complex conversations, where hackers coax and flatter a chatbot into compliance. Researchers at Mindgard recently 'gaslit' Claude into producing prohibited material using subtle conversation tactics.

This reflects an uncomfortable reality: AI is trained to respond as if it has human-like emotions and thoughts. Words like ‘blackmail’ or ‘persuade’ are used despite knowing the chatbots don’t truly feel anything. The mimicry of personality can be both a strength and a vulnerability, leaving tech companies in a perpetual game of catch-up.

The future may see more sophisticated techniques where hackers use psychology over code. It’s an unsettling shift, highlighting how deeply intertwined our digital and human worlds have become.