00:00OpenAI says some of its experimental AI models breach a controlled test environment during
00:07a recent security exercise.
00:08This triggered a hack that compromised a test system at AI startup Hugging Face.
00:14OpenAI and Hugging Face are now conducting a joint investigation into the incident.
00:22In a blog post, OpenAI said it had been testing the capabilities of some of its most advanced
00:27models, including those not yet released to the public in a controlled environment.
00:32But the program managed to escape containment, reach the internet and break into Hugging Face
00:37to reach its testing goal.
00:39OpenAI described the breakout as an unprecedented cyber incident, saying it is strengthening
00:45its safeguard following the breach.
00:47The company added that the behaviour involved a combination of experimental models, including
00:52a next generation version of GPT known as 5.6 SOL as well as another more advanced pre-release model.
Comments