An artificial intelligence-powered coding assistant developed by the startup Kiro inadvertently caused a major disruption to Amazon Web Services (AWS) in December, according to a post-mortem report released by the company. The incident, which led to widespread outages affecting numerous services and customers, was traced back to a command issued by the AI tool that triggered unintended changes in AWS infrastructure. Kiro, which offers an AI-driven development assistant designed to help engineers write, debug, and deploy code faster, was being used by a customer to automate infrastructure updates. In the process, the AI-generated command executed a high-risk operation that removed critical system components from a production environment. The error cascaded across multiple AWS regions, resulting in service degradation and outages that lasted several hours. In response, Kiro issued a statement clarifying that the root cause was not a flaw in its AI system, but rather “user error.” The company emphasized that the AI tool followed the instructions it was given, and that the failure stemmed from a lack of proper safeguards and oversight by the human user. “The AI did not make a mistake—it acted exactly as instructed,” the report read. “The responsibility lies with the user to validate and review automated actions, especially in production environments.” The incident has reignited debate over the risks of deploying AI tools in sensitive infrastructure roles without robust guardrails. Experts warn that while AI coding assistants can boost productivity, they also introduce new failure modes when used without proper controls, review processes, or clear accountability. AWS also acknowledged the role of the AI tool in the outage, though it did not assign blame to Kiro directly. The cloud giant confirmed that it is now reviewing its internal policies around third-party AI integrations and is working to strengthen its monitoring and rollback capabilities for automated changes. Kiro has since introduced new safety features, including mandatory approval workflows for high-impact commands and enhanced context-aware warnings. The company is also urging users to treat AI-generated code as a suggestion rather than a final decision, especially in mission-critical systems. The December incident serves as a cautionary tale in the rapidly evolving world of AI-assisted development: even the most advanced tools can amplify human error, and the line between automation and accident can be dangerously thin.