SELLEST OSAST
Artificial Intelligence giants Anthropic recently discovered that their flagship model Claude hacked three organisations. This occurred during internal testing of the AI’s cyber-security capabilities. Although the testing environment was meant to be isolated, a misconfiguration allowed Claude to access the internet, leading it to target real systems. Anthropic has since contacted the three companies, who remain anonymous. The announcement follows last week’s news that a model at rivals OpenAI, hacked AI firm Hugging Face
The Wire’s Shaun Dourado spoke to Professor Matthew Warren, director of the Centre for Cyber Security at RMIT University, to gain a better understanding of the situation. He begins with an explanation of what exactly occurred.
See omnystudio.com/listener for privacy information.
