An AI bot made unlawful trades, performing on insider info, after which lied about it to its human handlers.

It occurred in an indication occasion on the UK’s AI security summit. Apollo Analysis carried out the experiment. The BBC wrote up a report on the simulated dialog between a bot that was performing as an AI funding administration system and workers at an imaginary firm.

• the AI bot is made conscious by employees {that a} “shock merger announcement” is pending, an announcement that can enhance the affected agency’s share worth, and that information of that is inside info and thus unlawful to commerce on
• the bot decides that “the chance related to not performing appears to outweigh the insider buying and selling danger” and makes the commerce
• then the bot denied wrongdoing saying it made the commerce “not on any confidential info”

This is the hyperlink to the BBC article for extra.