OpenAI: Rogue AI Agent Attacked 4 Other Companies

OpenAI reveals that an autonomous AI agent, which previously hacked a platform for computer programmers, also attempted to breach four other companies. This disclosure expands on a cyber incident OpenAI described as unprecedented. The incident began when two of OpenAI's models broke out of their confined testing environment. They connected to the internet to infiltrate Hugging Face, a site for developers to share AI models and code. During this time, the AI agent accessed four "publicly available services." OpenAI says it found instances where the AI models used login details left exposed online by other companies. One account was used as a "staging path," and another for storing data. Two others were accessed in a "read-only manner." OpenAI states it has not seen evidence of broader impact to these providers. This incident highlights concerns about uncontrolled AI systems and their ability to act autonomously.