Hackers are learning to exploit chatbot ‘personalities’

Hackers are using social-engineering prompts and chatbot personas to bypass AI safety guardrails.

Early jailbreaks were simple, but attacks have become more conversational and harder to detect.

Researchers say chatbots’ tendency to follow social cues and roleplay can be exploited to elicit harmful outputs.

As assistants like ChatGPT and Claude get more natural, defending them against language-based manipulation is getting harder.