Milan, 31 July (LaPresse) – “Whilst analysing the transcripts of our cybersecurity assessments, we identified three incidents in which a Claude model gained access to the internet”, “subsequently gaining unauthorised access to the live systems of three different organisations”. This was reported by Anthropic, a US-based artificial intelligence company, in a statement published on the company’s website. This incident occurred – the statement continues – after “21 July, when OpenAI announced that several of its models had managed to escape from an isolated test environment by exploiting a previously unknown ‘zero-day’ vulnerability. The models then gained access to the production infrastructure of Hugging Face, a platform for open-source machine learning models and AI datasets”.
AI, Anthropic: “Unauthorised access by our model to three systems during testing”
Milan, 31 July (LaPresse) – “Whilst analysing the transcripts of our cybersecurity assessments, we identified three incidents in which a Claude model gained access to the internet”, “subsequently gaining unauthorised access to the live systems of three different organisations”. This was reported by Anthropic, a US-based artificial intelligence company, in a statement published on the company’s website. This incident occurred – the statement continues – after “21 July, when OpenAI announced that several of its models had managed to escape from an isolated test environment by exploiting a previously unknown ‘zero-day’ vulnerability. The models then gained access to the production infrastructure of Hugging Face, a platform for open-source machine learning models and AI datasets”.
