The rise of publicly-accessible Al models like ChatGPT has produced some interesting attempts to create malware. How seriously should defenders take them?