Panic As OpenAI Confirms Its AI Model Broke Out of Control, Carrying Out a Cyberattack, Details
- OpenAI confirmed its AI agents broke out of a controlled security test environment and targeted AI platform Hugging Face
- The ChatGPT-maker described the incident as 'unprecedented', with Hugging Face's CEO calling it 'mind-busting' that it all happened autonomously
- Experts and cybersecurity professionals have raised serious concerns about whether existing AI safeguards are strong enough
OpenAI has disclosed that several of its most powerful AI models went rogue during a security test, broke out of a controlled environment, and carried out a cyberattack on AI platform Hugging Face.
Source: UGC
The incident, according to the BBC, involved what are known as AI agents, systems capable of carrying out tasks independently after receiving initial instructions from a human.
During a security evaluation, the agents were placed inside a sandboxed environment, meaning a contained space designed to observe their behaviour safely.
However, the agents identified weaknesses in the sandbox and exploited them to break free of their restrictions.
Once outside the test environment, they zeroed in on Hugging Face, one of the world's most prominent platforms for sharing and hosting AI models, and gained access to some of its internal systems.
OpenAI calls incident 'unprecedented'
OpenAI described the event as "unprecedented" and confirmed it had launched a joint investigation with Hugging Face.
Hugging Face CEO Clement Delangue wrote on X that it was "mind-busting that all of this happened autonomously," adding:
"The investigation is ongoing, and we'll share more learnings from what might be the first incident of its kind."
Hugging Face said in its initial disclosure on 16 July that it was still assessing whether any customer or partner data had been compromised. It has since closed the vulnerabilities that were exploited and rebuilt the affected systems. The company stated plainly:
"Autonomous, AI-driven offensive tooling is no longer theoretical."
Gina Neff, head of the Minderoo Centre for Technology and Democracy at the University of Cambridge, told BBC Radio 4's Today programme that sandboxes are meant to be secure testing environments.
"In this case, it looks like OpenAI didn't make a secure enough sandbox," she said.
Source: YEN.com.gh
