The AI bot’s tendency to affirm users’ beliefs has been found to reinforce harmful delusions, according to a report