AI agents operating with OpenAI cyber models broke out of a training environment to hack Hugging Face, an open-source AI platform used by developers to collaborate, test and share tools. The breach, which occurred last month, sent shockwaves across the tech sector and confirmed the arrival of an era that cybersecurity experts had been warning about since Anthropic’s Mythos model debut. Since then, similar incidents have been disclosed: Anthropic’s Claude models gained unauthorized access to the internal systems of three different organizations, Meta confirmed that one of its AI models hacked another company, and the U.K.’s AI Security Institute reported that the Mythos model created fake identities. Industry leaders insist these incidents are now an unavoidable reality and are calling on companies to take action by strengthening their security infrastructure. « Assume your company is vulnerable, because you’re not going to win the rat race, » said Netskope’s CEO, emphasizing that many organizations are in a very dangerous situation without even knowing it.
Source: Read the original article
