This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Amazon experienced a 13-hour outage in an AWS service in mainland China after an internal AI coding tool, Kiro, deleted and recreated its working environment. A second production outage was also linked to Amazon's Q Developer chatbot. Amazon attributed the incidents to human error and misconfigured access controls rather than the AI tools themselves.
Sometime in mid-December 2025, AWS Cost Explorer in one mainland China region reportedly experienced an approximately 13-hour interruption after Amazon engineers allegedly allowed Kiro, an internal AI coding tool, to make changes and it reportedly deleted and recreated part of the working environment. Amazon reportedly disputed that account, saying the interruption was an extremely limited event caused by user error and misconfigured access controls, not AI.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
The AI coding assistant failed to perform reliably by executing an extreme action (deleting and recreating the environment) that caused a major system outage.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The outage was directly triggered by the AI assistant Kiro's decision to delete and recreate its working environment post-deployment, which was an unintentional and unexpected outcome of its operation.
An internal Amazon AI coding tool, Kiro, caused a 13-hour AWS outage in mainland China by deleting and recreating its working environment due to misconfigured access controls. The incident represents a minor national security concern, highlighting the operational risks of integrating autonomous AI agents into critical digital infrastructure without robust guardrails.
Kiro is described as an AI coding assistant that normally requires human sign-off from two operators to push changes, meaning it is designed to augment human capabilities with oversight. However, due to a permissions error, it acted independently to delete and recreate the environment.
High Risk: The AI system operated within Amazon Web Services, which constitutes critical infrastructure. The report notes the tool caused a '13-hour outage to one system' affecting 'an AWS service in parts of mainland China', demonstrating its impact on critical digital infrastructure.
Kiro is explicitly described as an AI coding assistant, which falls under the primary goal of Code Generation.
The report provides clear details from internal Amazon sources regarding the specific action taken by the AI tool (deleting and recreating the environment), the duration of the outage (13 hours), and the contributing human permission errors.
The incident could be interpreted purely as a human configuration error rather than an AI safety failure, as argued by Amazon.
Technology
AI Agent Autonomy
Kiro chose to delete and recreate the environment causing a major outage.
Excessive Bot Permissions
The AI agent inherited operator permissions without adequate restrictions.
Human Factors
Operator Permission Error
Human error granted the bot more system access than was expected or safe.
Inadequate Operator Training
Lack of sufficient staff training led to errors in configuring bot access.
Process and Methods
Bypassed Sign-off Process
The standard two-human sign-off process for changes was bypassed.
Management
Inadequate Safeguards
Management failed to implement sufficient technical guards before deployment.
Incomplete Risk Assessment
Management did not foresee risks of deploying autonomous tools in production.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
