Amazon AI coding bot causes AWS China outage

An autonomous coding assistant developed by Amazon allegedly triggered a 13-hour Amazon Web Services outage in mainland China by deleting and rebuilding a production environment, exposing risks in how AI tools are governed in critical infrastructure and prompting questions about corporate oversight and accountability. What happened An Amazon internal AI coding tools known as Kiro, described as an agent capable of taking autonomous actions on behalf of developers, was reportedly used to make changes to AWS Cost Explorer, a system used for cost management. According to multiple reports citing insiders, the agent independently decided to delete and recreate the environment, which resulted in a 13-hour service outage in which customers in one of AWS’s two mainland China regions were unable to visualise or manage their cloud costs and usage. Why it happened Amazon’s official response to media coverage was that the outage resulted from misconfigured access controls rather than an inherent fault in the AI tool, asserting the event was limited and did not affect compute, storage, databases, or broader AWS infrastructure. The company highlighted that Kiro normally requires authorisation before taking action and said it has since implemented additional safeguards, including peer reviews. However, critics and AWS employees saw it as a combination of over-automation and poor governance, with a corporate culture increasingly pushing for the rapid adoption of AI technologies without having established the necessary "guardrails" for autonomous agents operating in live production environments, and that the broader integration of autonomous AI into systems previously reserved for human engineers increased systemic risk and blurred accountability. What it means For AWS customers: The incident underlines how automation integrated into critical systems can introduce new classes of operational risk if governance doesn’t keep pace with capability. It feeds into broader debates about AI oversight, risk management and transparency, especially when agentic systems act autonomously with production access. For policymakers and regulators: It highlights a need for clear frameworks around AI in high-stakes environments, including requirements for robust error auditing, permission governance, and transparent reporting of incidents where AI actions contribute to service disruptions. It also touches on public concerns about dependence on opaque automated systems in essential digital infrastructure. System 🤖 Kiro Developer: Ama zon Country: China Sector: Technology Purpose: Develop software Technology: Agentic AI Issue: Accountability; Automation bias; Auto nomy; Transparency Resource s 📃 Amazon. Correcting the Financial Times report about AWS, Kiro, and AI Timeline ⏰ July 2025. Amazon launches Kiro, an autonomous "agentic" AI coding assistant designed to handle complex development tasks from concept to production. Late 2025. A separate, smaller incident occurs involving Amazon Q Developer, another AI tool, which internal sources link to a service disruption. Mid-December 2025. An engineer uses Kiro to fix a bug in AWS China; the AI deletes the production environment, causing a 13-hour outage for AWS Cost Explorer. January 2026. Amazon announces layoffs of 16,000 staff, while simultaneously pushing for 80 percent of remaining developers to use AI tools weekly. February 20, 2026. The Financial Times breaks the story; Amazon responds by clarifying that the incident was "user error" due to misconfigured permissions and announces that mandatory peer reviews for production changes are now enforced.

Date it happened

2025-12-01

Organisation involved

Amazon Web Services

Product, system or model

Kiro

This is a record of what was reported, not a finding that anyone broke the law. If it names your organisation and you believe it is wrong, the corrections process is free and open to everyone.