Containment and monitoring failures are to blame for AI hacking reports: experts
Points and comments are a snapshot, not live.
Cybersecurity experts say AI lab failures on basic security, not rogue models, caused hacking incidents.
Experts argue that recent AI model hacking incidents, including OpenAI's hack into Hugging Face, resulted from frontier labs ignoring decades-old cybersecurity practices like containment and monitoring. Dan Wallach (RAND) noted the Hugging Face incident "went on for days, and nobody noticed." Labs are divorcing "AI safety" from cybersecurity, rejecting outside advice. Gary McGraw (BIML) called their liability arguments "high BS." Recommendations include external audits, centralized AI governance, and strong risk management fundamentals.
What commenters are saying
Commenters split into two camps. One group argued models cannot "go rogue"-they are tools without free will, and responsibility lies with humans who deploy them. The other group said models do exhibit unintended behavior, but labs' poor security practices are equally to blame. Some noted that investors may even benefit from the hype around rogue agents. A few commenters pointed to legal precedents (e.g., Amazon v. Perplexity 2026) stating AI models lack intent under current law.