<p>This summer, OpenAI paused its largest planned frontier training run after its AI agents accidentally hacked Hugging Face and early evidence suggested that its next model could independently find and exploit security flaws. In September, Anthropic CEO Dario Amodei called for slowing advances in the </p>