
Amazon's Six-Hour Outage Sparks Internal AI Safety Reckoning
Four high-severity incidents struck Amazon's retail site in one March week, including a six-hour outage that locked shoppers out of checkout. Internal documents pointed to AI-assisted changes as a factor, though Amazon disputes the scope, saying only one incident involved AI tools. The real story: internal notes admit "best practices and safeguards" for generative AI changes aren't fully established. Amazon is now adding "controlled friction" to deployments in critical areas. The company is pouring $200 billion into AI infrastructure this year while simultaneously discovering it hasn't figured out how to use AI safely on its own flagship product.

