AWS blames user error, not AI, for cloud outage caused by AI

Agent decided to delete and recreate an entire environment

Amazon Web Services suffered two outages in December related to errors involved with its own AI tools – but has put the blame on employees, not technology.

"People familiar with the matter” told The Financial Times that a 13-hour outage affecting AWS at the end of last year was caused by Amazon's internal AI coding assistant, Kiro.

Kiro is an agentic system, apparently designed to do more than vibe code and create production-ready software.

Normally Kiro requests authorisation before taking an action. However, in the December incident it had been given permissions similar to a human engineer, and the change – which was meant to be a small fix to a customer-facing system – progressed without approval.

The bot apparently decided that the best way to fix the problem was a scorched earth policy, deleting and recreating the entire environment.

That triggered a chain reaction in one of AWS’ two regions in mainland China, causing the 13-hour outage. This affected a single service, AWS Cost Explorer, which helps customers visualise, understand, and manage AWS costs and usage over time.

The incident has apparently raised questions inside AWS about the freedom given to AI tools. But, AWS has also defended Kiro and told Reuters the “brief event” was the result of “user error - specifically misconfigured access controls - not AI.”

It said a human could have made a similar mistake.

However, the FT’s sources among AWS engineers said this the second time in recent months that an AI coding tool had been linked to a service issue - issues that were small but "entirely foreseeable".

The December incident, which mainly affected China, was unrelated to the global AWS outage in October, which affected banks, social media platforms, gaming services and Amazon’s own tools like Amazon Prime Video, Ring and Canva.

An AWS spokesperson contacted Computing post-publication of this article with the following statement:

“The disruption was an extremely limited event last December affecting a single service (AWS Cost Explorer) in one of our 39 Geographic Regions around the world. It did not impact compute, storage, database, AI technologies, or any other of the hundreds of services that we run.

“The issue stemmed from a misconfigured role - the same issue that could occur with any developer tool (AI powered or not) or manual action. We did not receive any customer inquiries regarding the interruption. We implemented numerous safeguards to prevent this from happening again - not because the event had a big impact (it didn't), but because we insist on learning from our operational experience to improve our security and resilience. Additional safeguards include mandatory peer review for production access.

“While operational incidents involving misconfigured access controls can occur with any developer tool - AI-powered or not - we think it is important to learn from these experiences. The Financial Times' claim that a second event impacted AWS is entirely false.”