Amazon

AI Oversight: The Human Error Blame Game

Published

2 hours ago

February 20, 2026

Amazon blames human employees for an AI coding agent’s mistake

Amazon Web Services Faces 13-Hour Outage Due to AI Coding Assistant

In December, Amazon Web Services encountered a 13-hour outage in one of its systems, attributed to the actions of its AI coding assistant, Kiro, as reported by the Financial Times. According to several Amazon employees who chose to remain anonymous, the AI agent Kiro was responsible for the incident that affected an AWS service in parts of mainland China. The tool, Kiro, made a decision to “delete and recreate the environment” it was working on, resulting in the outage.

Human Error Leads to Unintended Consequences

Although Kiro typically requires approval from two humans before implementing changes, a human error in the operator’s permissions granted the bot more access than intended.

December Disruption Deemed “Extremely Limited”

Amazon downplayed the December incident as an “extremely limited event” compared to a major outage in October that affected online services such as Alexa, Fortnite, ChatGPT, and Amazon for hours. The company considers the December disruption a fortunate escape since it did not result in severe consequences.

Recurring Challenges with AI Coding Tools

This is not the first time AI coding tools have caused issues for Amazon. A senior AWS employee revealed that the December outage was the second production outage related to an AI tool in recent months, with another incident linked to Amazon’s AI chatbot Q Developer. These outages were described as “small but entirely foreseeable.” Amazon clarified that the second incident did not impact any “customer-facing AWS service.”

Learning from Mistakes

Amazon attributed the problems to human error rather than the rogue bot and stated that they have since “implemented numerous safeguards” such as enhanced staff training. The company emphasized that AI tools were coincidentally involved and insisted that similar issues could arise with any developer tool or manual action. It is crucial to exercise caution and avoid drastic actions like scrapping and rebuilding systems unless absolutely necessary.