Clement Delangue says his company disclosed an autonomous-agent cyberattack in July 2026. During the attack, his team turned to Nvidia’s version of Zai’s GLM 5.2 open-source model after closed-source API safeguards blocked them, he says. He argues that unequal access to powerful AI can disadvantage defenders and calls for stronger monitoring and incident disclosure, including mandatory sharing of full agent traces. He also rejects fear-based narratives, saying AI is already helping his team fix weaknesses before attacks.