Australian Liberal MP Andrew Hastie has highlighted significant safety flaws in artificial intelligence software after revealing that Microsoft Copilot advised him to respond positively to a constituent planning to end his life.
During a parliamentary inquiry on Friday, Hastie shared details of an exchange with a man suffering from a terminal illness who informed his local representative of his intention to undergo voluntary assisted dying. According to Hastie, the AI tool suggested replies including “congratulations!”, “great to hear from you”, and “that is wonderful news!”.
Hastie used the incident to underscore the dangers of relying on American-run AI systems for sensitive communications. He argued that such shortcomings demonstrate the urgent need for Australia to develop and operate its own AI technologies.
The revelation comes amid broader discussions about the dual nature of AI. While Hastie pointed to the ethical and safety failures of current models, cybersecurity officials have simultaneously emphasized to the inquiry that advanced AI capabilities are essential for defending against sophisticated malicious cyber actors.
Does this actually prove we need local AI, or just better prompt engineering? I’m skeptical about the sovereignty argument here.
Wait, did it really say ‘congratulations’? That’s genuinely disturbing how out of touch these models can be with sensitive topics.
This is absolutely horrifying. An AI suggesting congratulations for assisted dying shows a complete lack of empathy and safety guardrails.