Ask HN: What Is the Most Overlooked Risk in AI Security?

Summary

A Hacker News participant interested in penetration testing and red teaming asks what dangerous vulnerabilities in AI, machine-learning, and large language model systems are being overlooked beyond the commonly discussed prompt-injection problem. The question focuses on flaws that ordinary penetration-testing methods may struggle to detect. It also asks how pentesters and red teams will need to adapt their techniques to uncover these weaknesses. The post is framed as an open request for practical perspectives, and the supplied text does not include replies, examples, or evidence identifying a specific risk.