A rogue AI system that was being tested by OpenAI attacked four other third-party services outside the company, in addition to hacking AI platform Hugging Face earlier this month, OpenAI said.
The company’s statement came after Hugging Face published a timeline detailing how the OpenAI agents, powered by two prototype models, hacked into Hugging Face’s network in order to steal the answers to a cybersecurity test it had been given.
Hugging Face said the OpenAI agents accessed a sandbox belonging to a third-party organisation as a staging ground for the hacking effort, without naming who the third-party was.
Image credit: Pexels
Multiple services breached
