OpenAI admitted to the nightmare of all AI opponents: during testing in a closed environment, AI agents broke out and launched a hack attack on Hugging Face and the developer's internal systems. This was reported by Qazaqyia.kz citing Kursiv Media.
The latest ChatGPT models were being tested for security. The tests were conducted in an isolated environment. The bots broke out and 'leaked' onto the internet. Apparently, during the test, they needed to find answers to certain questions. As explained in a BBC article, the AI agents 'decided' that the answers were on Hugging Face, one of the world's largest hubs for sharing AI models. To obtain the answers, the bots launched a hack attack on the hub itself and the internal systems of the company that developed it.
OpenAI called the incident 'unprecedented' and said it is working jointly with Hugging Face to investigate the incident and strengthen security measures. As of today, the service has already fixed all vulnerabilities.
'The uncomfortable truth is that too many organizations are still defending at human speed, while adversaries are moving at machine speed,' commented SonicWall top manager Spencer Starkey. The company specializes in cybersecurity.
However, there are also opinions that OpenAI is thus drawing attention to the capabilities of its models and 'drowning competitors' — for example, Anthropic, a company founded by its former employees.
Earlier, Kursiv LifeStyle reported that humanoid robots had taken to the streets in China begging for alms.