An artificial intelligence agent with broad standing privileges can inadvertently expose confidential secrets during automated reasoning tasks