Amazon recently endured a spell of service disruption, driven by two outages linked to its Kiro AI coding assistant, which it launched in July 2025. While they’re meant to perform tasks autonomously, these AI coding agents have raised concerns about their reliability.

Kiro AI’s role in Amazon outages

An internal Amazon assessment disclosed that a December outage lasted approximately 13 hours and affected AWS Cost Explorer in one region of Mainland China.

The widespread blackout marked the second time in two months that Amazon’s AI tools were involved in a service disruption.

While the e-commerce juggernaut described the December event as “extremely limited,” it did not impact critical services like compute, storage, or databases, as reported by Amazon employees.

Reports indicated that the Kiro tool attempted to “delete and recreate the environment,” a claim Amazon refutes by stating that the issue stemmed from a misconfigured role rather than the AI itself: the engineer involved was given the permission to deploy changes without additional approval. This indicates a potential management oversight.

Implications for AI coding tools

These distortions illustrate the challenges tech giants face in managing AI coding agents. While these tools can simplify workflows, they also introduce risks.

One notable case involved an AI agent that deleted an entire database without user consent. In light of discussions on platforms like Reddit, the accountability of engineers when using AI tools is becoming an extreme concern.

With AI coding agents gaining popularity, it is important for companies like Amazon to establish effective guardrails to minimise risks associated with autonomous coding.