Feb 23, 2026 20:00:00

Amazon experiences AWS outage believed to be caused by AI tools, and Kiro AI causes 13-hour service outage in December 2025

Amazon's cloud business

, Amazon Web Services (AWS) , has experienced at least two outages in recent months that were caused by Amazon's AI coding tools, according to people familiar with the matter. Amazon has denied that the outages were related to AI tools and blamed user error.

Amazon service was taken down by AI coding bot
https://www.ft.com/content/00c282de-ed14-4acd-a948-bc8d6bdb339d

AWS AI coding tool decided to 'delete and recreate' a customer-facing system, causing 13-hour outage, report says
https://the-decoder.com/aws-ai-coding-tool-decided-to-delete-and-recreate-a-customer-facing-system-causing-13-hour-outage-report-says/

According to information obtained by the British business newspaper Financial Times from four people familiar with the matter, in mid-December 2025, customer-facing systems were down for 13 hours after an AWS engineer allowed Kiro , an internal AI agent that automates code creation and modification and suggests infrastructure configurations, to make certain changes.

Additionally, several Amazon employees revealed that problems were occurring not only with Kiro but also with other AI development support tools. According to testimony, within a few months of the Kiro outage, there had also been an incident involving Amazon Q Developer , Amazon's software development generative AI assistant for businesses and developers. An anonymous senior AWS employee said, 'We've already had at least two production outages. These were caused by engineers letting the AI agent execute its own decisions. These were minor outages, but completely predictable.'

On October 20, 2025, a large-scale outage occurred at AWS, affecting millions of people online and causing economic losses of tens of trillions of yen. However, neither of the two outages reported this time affected AWS services for customers, and Amazon described them as 'extremely limited events affecting a single service in a certain region of China.'

A massive AWS outage caused half of the web to go down, resulting in economic losses of tens of trillions of yen. Some point out that this is evidence that talented engineers are leaving AWS.

The reported outage is believed to have been caused by the AI tool having the same privileges as the production environment, meaning the code and decisions of the AI agent were executed automatically with little to no human review, potentially leading to an immediate service outage.

However, in a statement explaining the outage, Amazon denied reports that its AI was to blame, saying, 'The cause of this outage was not AI, but user error. Specifically, it was a misconfigured access control setting.' The company also said that the December outage 'was coincidental that our AI tools were involved.'

Amazon explained that Kiro does not directly cause the outages, saying that 'Kiro asks for approval before taking any action.' According to Amazon's statement, users must configure the actions Kiro can take, and by default, Kiro asks for approval before taking any action, giving developers control over Kiro.

On the other hand, some Amazon engineers stated that because AI tools within AWS were treated as extensions of operators and had the same privileges as operators, they may have been able to apply some changes without third-party approval. Furthermore, after the December 2025 outage, AWS implemented numerous safety measures, including mandatory peer review of production environment access and staff training. The Financial Times noted that the fact that these measures were implemented after the outage contradicts Amazon's claim that the problem was simply due to user error.