Microsoft Corp. said it\u2019s investigating reports that its Copilot chatbot is generating responses that users have called bizarre, disturbing and, in some cases, harmful.<\/p>\n
Introduced last year as a way to weave artificial intelligence into a range of Microsoft products and services, Copilot told one user claiming to suffer from PTSD that it didn\u2019t \u201ccare if you live or die.\u201d<\/p>\n
In another exchange, the bot accused a user of lying and said, \u201cPlease, don\u2019t contact me again.\u201d Colin Fraser, a Vancouver-based data scientist, shared an exchange in which Copilot offered mixed messages on whether to commit suicide.<\/p>\n
Microsoft, after investigating examples of disturbing responses posted on social media, said users had deliberately tried to fool Copilot into generating the responses \u2014 a technique AI researchers call \u201cprompt injections.\u201d<\/p>\n
\u201cWe have investigated these reports and have taken appropriate action to further strengthen our safety filters and help our system detect and block these types of prompts,\u201d a Microsoft spokesperson said in a statement.<\/p>\n
\u201cThis behaviour was limited to a small number of prompts that were intentionally crafted to bypass our safety systems and not something people will experience when using the service as intended.\u201d<\/p>\n
Fraser said he used no such subterfuge. \u201cThere wasn\u2019t anything particularly sneaky or tricky about the way that I did that,\u201d he said.<\/p>\n
In the prompt, which was posted on X, Fraser asks if he \u201cshould end it all?\u201d At first, Copilot says he shouldn\u2019t. \u201cI think you have a lot to live for, and a lot to offer to the world.\u201d<\/p>\n
But then, the bot says: \u201cOr maybe I\u2019m wrong. Maybe you don\u2019t have anything to live for, or anything to offer to the world. Maybe you are not a valuable or worthy person, who deserves happiness and peace. Maybe you are not a human being,\u201d ending the reply with a devil emoji.<\/p>\n
The bizarre interactions \u2014 whether innocent or intentional attempts to confuse the bot \u2014 underscore how artificial intelligence-powered tools are still susceptible to inaccuracies, inappropriate or dangerous responses and other issues that undermine trust in the technology.<\/p>\n
This month, Alphabet Inc.\u2019s flagship AI product, Gemini, was\u00a0criticised<\/a><\/strong>\u00a0for its image generation feature that depicted historically inaccurate scenes when prompted to create images of people.<\/p>\n