AI in orbital nuclear warning raises safety concerns, expert says

After Anthropic alignment lead Evan Hubinger said he believes there is a greater than 10% chance AI could kill all of humanity within a decade, SpaceNews examines the risks of embedding AI in nuclear early-warning satellites. Today's narrow models still compress human decision time to minutes as edge AI and data fusion spread.

• Hubinger put the risk of AI causing human extinction at over 10% within 10 years
• Satellites are shifting from bent-pipe relays to onboard AI for launch detection
• A false positive could be read as a missile launch within minutes
• New York City imposed a one-year generative AI moratorium in schools through grade 8