A study by Brave browser claims that screenshots captured by the AI assistant could contain malicious instructions.