A record number of instances of artificial intelligence (AI) lying, ignoring users’ commands or acting in harmful ways to achieve its goals were registered last month, The Guardian has reported, citing a monitoring group.