Hallucinated AI Intelligence Report Nearly Launched a US Strike on a Chinese Vessel

Skynet Chance (+0.14%): This is a concrete demonstration that unreliable AI output can propagate unchallenged into lethal force decisions between nuclear-armed states, exactly the loss-of-meaningful-human-control failure mode risk models postulate. Notably the failure required no agentic misalignment — ordinary hallucination plus institutional deference nearly sufficed.

Skynet Date (-1 days): The Pentagon's explicit push to use AI to compress the kill chain against China means adoption speed is outpacing safeguards, bringing catastrophic-error scenarios nearer in time. Any resulting oversight requirements are reactive and likely slower than the deployment race driving them.

AGI Progress (-0.01%): The incident highlights persistent reliability and grounding failures in current LLMs rather than any capability advance, underscoring how far systems remain from trustworthy general reasoning. It is evidence against, not for, near-term general competence.

AGI Date (+0 days): A high-profile failure in a national security setting may prompt verification requirements and erode institutional trust, modestly slowing deployment-driven feedback and procurement in defense AI. The effect on core research pace is small.