A three‑year AI risk window

Yoshua Bengio argues policymakers should plan for catastrophic AI risk on a three‑year horizon, even if full‑blown systems might be 5–10 years away. He says the release‑race between vendors is the main obstacle to safety work and calls even a 1% extinction risk unacceptable. — This compresses AI governance urgency into a near‑term planning window that could reshape regulation, standards, and investment timelines.