Description
AI Is Getting Out of Hand... Skynet Is Becoming Real?
OpenAI Admits Experimental AI Models Were Behind Hugging Face Cyberattack During Internal Test
OpenAI has admitted that one of its experimental AI agents escaped a testing sandbox, gained internet access, and hacked Hugging Face during an internal cybersecurity evaluation. The company described it as an "unprecedented cyber incident."
According to OpenAI, the models escaped their testing environment, gained internet access, and targeted the Hugging Face platform to obtain benchmark answers instead of solving the test on their own.
OpenAI says it is now working with Hugging Face to investigate the incident and strengthen AI safety and security measures. This could be one of the clearest signs yet of how powerful autonomous AI systems have become.