Four crashes in a week: AI-assisted changes broke Amazon's retail site

In one week of March 2026, Amazon's retail site suffered four high-severity incidents, including a six-hour meltdown that locked shoppers out of checkout, account information and product pricing. Internal documents prepared for a 'deep dive' run by the SVP overseeing Amazon's ecommerce infrastructure initially identified 'GenAI-assisted changes' as a factor in a pattern of incidents stretching back to Q3 — a reference that was deleted before the meeting took place, according to the Financial Times.

Amazon's blog post disputed the framing: it said only one incident involved AI tools, that 'none of the incidents involved AI-written code,' and pinned the cause on 'an engineer following inaccurate advice that an agent inferred from an outdated internal wiki.' Either way, SVP Dave Treadwell told staff that best practices and safeguards around generative AI usage hadn't been fully established, and that Amazon would introduce 'controlled friction' into deployments involving the most critical parts of the retail experience.

The reckoning landed at an awkward moment: Amazon had just surpassed Walmart to top the Fortune 500, is spending a projected $200 billion on AI infrastructure this year, and laid off roughly 14,000 corporate workers in October and another 16,000 in January in a push its CEO framed around AI-driven efficiency. The message to engineers after the outages: AI-assisted changes now get more scrutiny, and humans move back into the loop.

An agent inferred deployment advice from an outdated internal wiki, and an engineer followed it into critical retail systems.

Safeguards for generative-AI usage had not kept pace with adoption — 'best practices and safeguards' were admittedly not fully established.

The AI push ran alongside deep layoffs and a 'lean, move fast' culture, thinning the human review layer as AI-assisted changes multiplied.

An AI agent is only as current as its sources: advice inferred from a stale wiki went straight into production — keep humans where failure is expensive

Amazon said it would put humans 'further back in the loop,' adding controlled friction to critical-path deployments and more scrutiny of AI-assisted changes. It disputed reports that it imposed new approval requirements for engineers using AI tools or that AWS was involved, and called the emergency-sounding review a routine weekly operations meeting. The deleted line about GenAI-assisted changes, though, became the story.