Illustration, Photo: Reuters
The American company OpenAI revealed last night that the first fully autonomous hacking attack by rogue artificial intelligence models involved more companies than initially thought.
Until now, it was believed that the Hugging Face platform was the only victim of the unprecedented hacking attack by rogue ChatGPT models, but OpenAI has now admitted that its bot attacked several other “publicly available services”, reports the BBC.
The company said that an out-of-control artificial intelligence system found login credentials for four user accounts online, which allowed it to access them.
Meanwhile, the company Hagging Faces, at an emergency meeting with hundreds of cybersecurity experts, described what it was like to be the target of the world’s first fully autonomous artificial intelligence hacking attack.
The company described how the artificial intelligence worked at superhuman speed, but at the same time made strange decisions and made mistakes that no human, like a hacker, would make.
The Hagging Face platform, which is like an app store for artificial intelligence tools, stated that the hacking models were trying out thousands of different methods simultaneously.
The company previously announced on July 16 that it had been targeted by a hacker using powerful autonomous artificial intelligence and that it had reported the case to the police.
Almost a week later, OpenAI admitted that its artificial intelligence had broken out of a closed environment during testing and independently attacked the Hagging Face platform.
This system targeted Hagging Face while he was trying to find answers to a hacking exam given to him by the company OpenAI.
Today, OpenEye updated its statement, revealing that the attack was more extensive than initially thought.
“The models identified and exploited publicly exposed login credentials for user accounts on other publicly available services. This includes four accounts on four services,” OpenEye said.
She did not specify whether she meant other companies by “publicly available services”, but stated that the new attacks were not of the same level of severity as the hacking of the Hagging Face platform.
On Tuesday, the industry body Cloud Security Alliance released a report based on an emergency meeting held by the company Hagging Face on Friday.
The alliance stated that the AI models “often chose inefficient paths and exhibited clumsy behavior that no human would choose.”
The AI models were repeating actions they had previously performed, which is a sign that the agent AI was losing task flow and context, the Alliance added.
The Hagging Face platform warned that the artificial intelligence models, despite errors and strange behavior, “performed brilliant technical moves” and were able to quickly adapt to new scenarios during the multi-day hacker attack.
The attackers were only discovered in the Hagging Face network after three days, and it took the company’s artificial intelligence and cybersecurity experts many hours to stop them and remove them from the system, which would have been a serious problem for ordinary companies.
The company did not disclose the financial damage it suffered from the hack, but said employees had worked long hours to restore a third of its infrastructure.
Download the app and follow the news
FOLLOW US ON
News
Click Here For The Original Source.
