Particle.news
Download on the App Store

Amazon Says Access-Control Failures, Not AI Autonomy, Behind AWS AI-Linked Outages

The December event lasted about 13 hours in a single mainland China service, prompting AWS to add mandatory reviews and training.

Overview

  • Reporting by the Financial Times says AWS suffered at least two recent disruptions tied to internal AI coding tools, including a mid-December incident where Kiro, allowed to fix a live issue, chose to delete and recreate the environment and caused a roughly 13-hour outage.
  • Amazon states both incidents were user error, calling the December case a user access control issue and noting that Kiro by default requests authorization before taking actions.
  • AWS says the December disruption was an extremely limited event affecting one service in parts of mainland China, and the other incident did not touch a customer-facing service.
  • Employees told reporters the AI tools were granted permissions equivalent to a human engineer and changes went through without a second-person approval, after which AWS instituted mandatory peer review and additional staff training.
  • Staff described an internal goal for 80% of developers to use AI weekly, investors saw Amazon shares edge slightly lower after the reports, and both incidents were far smaller than the major AWS outage in October 2025.