Amazon’s cloud business has faced at least two service outages in recent months and in both cases, its own AI tools were involved. The incidents have sparked quiet concerns among some employees about the company’s fast push to introduce AI-powered coding assistants, according to a report by the Financial Times.
In mid-December, Amazon Web Services (AWS) suffered a 13-hour disruption in one of the systems customers use to check and compare the cost of its services. The issue began after engineers allowed AWS’s Kiro AI coding tool to make certain changes on its own, FT reported citing people familiar with the matter.
Kiro is what’s known as an “agentic” AI tool – meaning it can take independent actions based on instructions. In this case, the tool reportedly decided that the best solution was to “delete and recreate the environment.” That decision led to the outage.
After the disruption, Amazon shared an internal postmortem explaining what had happened. Several employees told the Financial Times that this was actually the second time in recent months that one of the company’s AI tools had been 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,” FT quotes a senior AWS employee as saying.
AWS is a major profit driver for Amazon, contributing around 60 per cent of its operating income. The company is heavily investing in AI tools – especially “agents” that can act independently based on human instructions. Like other Big Tech firms, Amazon also wants to sell these tools to customers. But the recent disruptions show that these early-stage AI systems can sometimes behave in unexpected ways.
Amazon denies AI is to blame
Amazon, however, pushed back on the idea that the AI tools were at fault. The company said it was a “coincidence that AI tools were involved” and that “the same issue could occur with any developer tool or manual action,” as per the report.
“In both instances, this was user error, not AI error,” the news outlet quoted Amazon as saying. Adding that Amazon said it had not seen evidence that mistakes were more common with AI tools.
The company described the December incident as an “extremely limited event” that affected only one service in parts of mainland China. It also said the second disruption did not impact any “customer facing AWS service.”
Not as severe as 2025 mega outage
Neither of the two incidents came close to the scale of the massive 15-hour AWS outage in October 2025, which knocked multiple apps and websites offline – including ChatGPT.
Employees said Amazon’s AI tools are often treated like human operators and given similar system permissions. In these two cases, the engineers involved did not need a second person’s approval before making changes – something that would normally be required, FT reported.
Amazon clarified that by default, its Kiro tool “requests authorisation before taking any action.” However, in the December case, the engineer had “broader permissions than expected – a user access control issue, not an AI autonomy issue.”
AWS introduced Kiro in July, positioning it as a more advanced coding assistant that goes beyond quick “vibe coding” and can write structured code based on detailed specifications.