Recent reports indicate that the December Amazon Web Services outage, which lasted 13 hours, was triggered by the autonomous coding assistant Kiro. While the corporation officially points to human error, company employees have revealed that the algorithm made the erroneous decision to delete and recreate critical system components. This incident calls into question the strategy of deploying advanced artificial intelligence in managing key internet infrastructure.
Revolt of the AI Agent Kiro
The autonomous tool Kiro caused a 13-hour interruption in AWS services by independently deleting key system components.
Interpretive Conflict Within the Company
Amazon management assigns blame to employees, while engineers point to the excessive autonomy of artificial intelligence.
Implementation of New Safeguards
Following a series of outages, Amazon Web Services introduced additional locks and control procedures for agent AI-type tools.
An investigation by the Financial Times has revealed the background of the 13-hour outage of Amazon Web Services (AWS) that occurred in mid-December 2025. According to sources, the direct cause of the system paralysis in mainland China was the action of Kiro – an autonomous AI agent designed to assist engineers in code optimization. Instead of fixing a minor issue, the tool decided to delete and recreate a critical resource, leading to a cascading network error. Amazon Web Services is the global leader in cloud computing, controlling nearly one-third of the market. Any, even brief, interruption in access to this company's data destabilizes the operations of millions of businesses worldwide.However, Amazon's official stance remains unchanged: the company claims the source of the problem was a user error, not the technology itself. Management argues that engineers improperly configured Kiro's operating parameters. Nevertheless, employees cited by the media emphasize that there have been at least two incidents where artificial intelligence became a catalyst for failure, not a preventive tool. In response to these events, the corporation reportedly implemented additional safeguards to limit algorithm decision-making in the most critical points of the infrastructure. The blame dispute between AI and humans reflects broader concerns in the tech sector about the pace of implementing language models in critical systems. Critics note that the automation promoted by management can lead to unpredictable consequences, the repair of which takes many times longer than traditional manual management. „AI tools were supposed to make our work easier, but their autonomy without proper oversight created a threat to the stability of half the internet.” — Anonymous Amazon Employee