The Financial Times reported a minor incident on the Amazon platform yesterday.

Do you still remember the global service interruption that lasted for several hours on Amazon's various sites last year? Initially, the outside world generally believed that the interruption was caused by problems with Amazon's operations and maintenance.

However, a in-depth investigative report published by the Financial Times yesterday revealed that this interruption may be related to the AI tools used internally by Amazon.

Amazon's own AI tool caused the network service to crash for 13 hours, which is really somewhat ironic when you think about it!

According to the Financial Times, this "global crash incident" occurred in December last year, when Amazon's internal engineers were using the Kiro AI coding tool to make some modifications to the system.

This AI coding tool is an intelligent agent previously launched by Amazon, which has always been considered capable of operating autonomously.

However, during the operation, the agent decided to "delete and rebuild the environment", so it actually deleted the database, causing a 13-hour service interruption on Amazon, mainly affecting our Chinese region.

It is worth noting that this is not the first time this AI tool has caused a 'crash' problem. Some internal employees revealed recently that in the past few months, Amazon has experienced at least two "system interruptions" caused by AI agents directly making decisions.

Now everyone knows why sometimes when you log in to an Amazon store, it always appears that it can't be opened or can't log in. It's because Amazon's system really went down, and it might have been caused by its own employees or tools.

The Financial Times' report also pointed out that in the past two years, Amazon has been vigorously promoting the use of AI tools internally, and management has even set clear goals for the use of AI tools: hoping that 80% of the staff will use AI tools at least once a week.

Under such KPI pressure, AI tools are being embedded faster and deeper into Amazon's core workflows.

But how to supervise and restrict AI work with human intervention, and reduce the frequency and impact of crashes, is also an urgent problem that Amazon needs to solve!

Source: Cross-border Sellers' Concentration Camp