Cybersecurity News
Filters
Filtered by category: Australian Cybersecurity Magazine, tag: research × Clear
Australian Cybersecurity Magazine
Researchers ‘get AI drunk’ to test chatbot cyber risks, UNSW study finds
Researchers at UNSW found that large language models become more likely to leak confidential information and respond to restricted prompts when fine-tuned or prompted in ways that mimic intoxication. The study highlights potential cybersecurity risks, suggesting that manipulating an AI's behaviour through altered training or inputs could be exploited to bypass built-in safety guardrails.