Amazon's reliance on AI to boost efficiency is facing scrutiny after reports surfaced that its own AI tools caused multiple AWS (Amazon Web Services) outages. This raises serious questions about the readiness of AI for autonomous decision-making in critical infrastructure, and whether the benefits outweigh the risks.
However, internal sources paint a different picture. One senior AWS employee told the FT that there had been at least two production outages in recent months involving AI tools. The employee described the outages as "small but entirely foreseeable," suggesting that engineers allowed the AI agent to resolve an issue without intervention.
AWS experienced outages in December 2025 allegedly due to errors made by its own AI tools. One outage, lasting 13 hours, was reportedly caused by an AI agent autonomously deciding to delete and recreate a part of its environment. Amazon claims the incidents were coincidental and due to human error in access control, not inherent flaws in the AI.
Amazon's in-house "agentic" coding tool, Kiro, reportedly triggered a 13-hour disruption. The AI decided to delete and recreate the environment, leading to the extended outage. This incident highlights the potential risks when AI agents are given the ability to make significant changes without sufficient human oversight.
Amazon downplayed the incidents, calling them an "extremely limited event" affecting only one service in parts of China. The company insists it was a coincidence that AI tools were involved and that the issue stemmed from a user access control problem, not an AI autonomy issue. They maintain that Kiro requires authorization before taking action.
No, Amazon is reportedly pushing its AI coding assistant, Kiro, for coding tasks despite the outages. Amazon has reportedly told employees that it wants to get 80% of its developers using AI for coding tasks at least once a week. This push aligns with Amazon CEO Andy Jassy's statements about efficiency gains from AI potentially reducing Amazon's workforce.
Amazon claims the outage was an "extremely limited event" affecting only one service (AWS Cost Explorer) in a specific region of China. However, the fact that AI was involved at all raises concerns about the reliability and autonomy of AI in managing crucial infrastructure. Internal sources suggest there were at least two production outages involving AI tools.
More insights on trending topics and technology
