AI Agents Are a Security Ticking Time Bomb

“The more a system reasons, the more unpredictable it becomes.” These words from Ilya Sutskever, former chief scientist and co-founder of OpenAI, have been at the top of many people’s minds since his talk at a recent conference. He argued the AI industry had reached the limits of pre-training large language models (LLMs). He will now turn to creating superintelligent agents — systems capable of reasoning, understanding, and performing complex tasks.

While Sutskever warns that the next generation of AI agents will develop their own…