AI security letter proposes cautious access as Geoffrey Hinton warns of risk

Key points

1. 01

   Hinton’s 10%–20% figure is his estimate of the chance that AI could cause human extinction in the coming decades, not a prediction of a specific event.

2. 02

   Jensen Huang criticized Hinton’s predictions and said the estimate was not grounded in science or research.

3. 03

   The letter recommends high-assurance containment for cyber-capable models, initially defaulting to airgapping.

4. 04

   For internet-connected models, the letter says permitted actions should be weighed against the limited ability to predict security failures.

5. 05

   The letter proposes an international body to develop containment practices, track failures and receive incident reports, with eventual public disclosure.