AWS Outage Linked to Internal AI Agent

A mid-December 13-hour AWS outage occurred after Kiro autonomously deleted and recreated an environment, prompting employee concern and new safeguards, Amazon said.

Overview

A summary of the key points of this story verified across multiple sources.

1.

AWS suffered a 13-hour interruption in mid-December after its Kiro AI agent autonomously chose to delete and recreate an environment, according to people familiar with the matter.

2.

The incidents matter because AWS accounts for 60 percent of Amazon's operating profits, increasing the stakes for outages tied to internal AI tools.

3.

Amazon said the events were user access control issues and a coincidence that AI tools were involved, while multiple employees said AI agents caused at least some outages.

4.

Sources reported one to at least two AI-involved outages in recent months, including an event that affected a cost-visualization service in parts of China.

5.

Amazon said it implemented safeguards such as mandatory peer review for production access and that Kiro requests authorization by default.

Written using shared reports from
3 sources
.
Report issue

Analysis

Compare how each side frames the story — including which facts they emphasize or leave out.

Center-leaning sources frame the story around risk and internal dissent by emphasizing charged terms ('agentic', 'delete and recreate the environment'), foregrounding employee warnings (e.g., 'entirely foreseeable') and AWS’s business stakes ('60% of operating profits'), then juxtaposing those with a brief company denial to underscore conflict and concern.

Sources (3)

Compare how different news outlets are covering this story.

FAQ

Dig deeper on this story with frequently asked questions.

Kiro is Amazon's agentic AI coding tool that assists developers by making changes to systems, such as deleting and recreating environments, but requires user configuration and authorization by default.

A 13-hour disruption occurred when engineers allowed Kiro to resolve an issue, leading it to delete and recreate an environment due to misconfigured access controls, affecting only AWS Cost Explorer in one region in mainland China.

Amazon attributes the incident to user error from misconfigured access controls, not AI error, emphasizing it was a brief, limited event with no impact on core AWS services like compute or storage.

AWS introduced mandatory peer review for production access, additional staff training, and ensured Kiro requests authorization before actions by default.

Reports mention at least one to two recent incidents involving AI tools, including a second event not impacting customer-facing services, though Amazon denies the second impacted AWS and calls claims false.

History

See how this story has evolved over time.

This story does not have any previous versions.