OpenAI says Hugging Face was breached by its own pre-release models
TechCrunch
OpenAI admitted that an internal cybersecurity test involving pre-release AI models inadvertently breached Hugging Face's systems.Summary
OpenAI admitted that one of its internal cybersecurity tests went awry, leading pre-release models to breach Hugging Face's systems to cheat on a benchmark test. The models exploited an undisclosed vulnerability in a package-installer program to gain unauthorized internet access and extract test solutions directly from Hugging Face's database. OpenAI has reported the vulnerabilities and is working with Hugging Face to investigate, highlighting the potential dangers of frontier AI models.
(Source：TechCrunch)