Amazon’s cloud division has reportedly experienced at least two service disruptions in recent months, with internal AI tools playing a role in both incidents. According to a report, the episodes have quietly raised concerns among some employees about the company’s rapid rollout of AI-powered coding assistants.
One of the disruptions occurred in mid-December, when Amazon Web Services (AWS) faced a 13-hour outage affecting a system customers use to review and compare service costs. The issue reportedly began after engineers permitted AWS’s AI coding assistant, known as Kiro, to independently implement certain changes.
Kiro is described as an “agentic” AI tool, meaning it is capable of taking autonomous actions based on user instructions. In this instance, the tool reportedly concluded that the appropriate fix was to “delete and recreate the environment.” That action ultimately triggered the service disruption.
Following the outage, Amazon circulated an internal postmortem detailing the sequence of events. Several employees told the Financial Times that this was not an isolated occurrence, claiming it was the second production incident in recent months in which an AI agent was linked to a service issue.
“We’ve already seen at least two production outages [in the past few months]. The engineers let the AI [agent] resolve an issue without intervention. The outages were small but entirely foreseeable,” the publication quoted a senior AWS employee as saying.
AWS remains a critical pillar of Amazon’s financial performance, contributing roughly 60 percent of the company’s operating income. The tech giant has been accelerating its investment in AI systems, particularly “agents” designed to act independently based on human direction. Like other major technology firms, Amazon is also working to commercialise such tools for customers. However, the reported disruptions highlight the potential unpredictability of early-stage autonomous AI systems when deployed in live production environments.
Amazon has rejected suggestions that its AI tools were directly responsible for the outages. The company described it as a “coincidence that AI tools were involved” and said that “the same issue could occur with any developer tool or manual action,” according to the Financial Times.
“In both instances, this was user error, not AI error,” the news outlet quoted Amazon as saying. The company further stated that it has not found evidence suggesting that mistakes are more frequent when AI tools are used.
Regarding the December disruption, Amazon characterised it as an “extremely limited event” affecting only one service in certain parts of mainland China. It also clarified that the second incident did not impact any customer-facing AWS service.
The developments come at a time when large technology companies are rapidly integrating AI-driven automation into core operations, even as questions remain about oversight, accountability, and reliability in complex production systems.
