Researchers at Permiso warn that threat actors can plant phishing messages within Copilot AI summaries. Notably, the researchers found that attackers can trick Copilot into including internal information to craft a more targeted message.