
When the AI tool Sybil flagged a previously unknown and complex vulnerability in a customer’s system last November, its creators at the startup RunSybil knew they were witnessing more than just a clever hack. They described it as a ‘step change’ in machine reasoning – a leap forward in an AI’s ability to understand intricate, interacting systems. This single event is a powerful illustration of a much broader trend. According to UC Berkeley computer scientist Dawn Song, we have reached a critical ‘inflection point.’ The rapid advancements in models are drastically increasing AI’s hacking prowess. This surge in general AI capabilities, as discussed in ‘Google Gemini Powers Apple’s Siri & New AI Features’ [8], is now creating a dual-use dilemma...








