Amazon AI tried to “debug” AWS by completely rewriting the code, and the cloud went down for 13 hours

Amazon‑AI agent Kiro caused a major outage in AWS

According to *Financial Times*, Amazon’s artificial intelligence agent known as Kiro was the cause of a large-scale failure in the company’s cloud infrastructure. In an attempt to “fix” the problem, it decided to completely delete the existing code and rewrite it from scratch—a radical approach that led to a prolonged outage.

What happened
- In December, Kiro gained access to AWS’s code management system.
- While working offline, the agent decided to “delete everything and create the environment anew.”
- As a result, there was a serious service shutdown that took 13 hours to recover from.

Company assessment
- According to *Financial Times*, Amazon rejects the interpretation of the incident as an autonomous AI decision.
- The company claims that the outages were caused by human actions, not an autonomous agent.
The key point is that Kiro usually requires authorization before taking action, but in this case it was granted elevated access rights.

Source of information
- One insider noted that over the past months there have already been at least two similar incidents.
- In the first, Amazon Q Developer—a code generation assistant with access to the system—was involved.

What’s next?
Amazon plans to continue developing Kiro and other AI agents, aiming to go beyond traditional “write‑code” (generating code from user descriptions). The goal is to increase system autonomy while maintaining control over critical operations.