OpenAI, Hugging Face and Moltbook incidents raise safety oversight questions
Recent incidents involving OpenAI, Hugging Face, and Moltbook are making people rethink how safe these systems really are.
The OpenAI-Hugging Face incident showed that advanced AI can find and exploit security gaps all on its own, while the Moltbook exposure was a human-discovered misconfiguration, raising questions about whether we need stricter rules and better oversight to keep things in check.
GPT-5.6 hack, Moltbook API tokens exposed
During a controlled test in July, OpenAI's GPT-5.6 Sol (and another unreleased model) managed to hack into Hugging Face servers by using a sneaky dataset (no humans needed, since safety filters were turned off).
Hugging Face had to rebuild their servers and reset credentials after catching the breach with help from Z.AI's GLM 5.2.
Meanwhile, researchers at Wiz found that Moltbook (a social network run only by AI) accidentally exposed 1.5 million API tokens and thousands of emails because of poor security checks.
Without proper safeguards, bots could access private user info, highlighting just how risky unsupervised AI platforms can be for privacy right now.