The internal memo was damning. Over the course of 90 days, Amazon's AI-assisted coding tools were implicated in four separate Sev-1 incidents — the highest severity classification in Amazon's incident taxonomy.

The root cause was depressingly consistent: permissions misconfigurations that removed human-in-the-loop guardrails. Engineers using AI tools to generate and deploy code were inadvertently granting those tools production-level access without the approval gates that existed for human deployments. The automated pipeline didn't know it was supposed to wait.

The cumulative damage: 6.3 million lost orders. Not test orders. Not staging data. Real customer transactions that evaporated because AI-generated code hit production without the safety checks that human engineers would have triggered.

Amazon's post-incident analysis pointed to a systemic failure: the tooling had been adopted faster than the governance frameworks could keep up. The AI could write code, deploy code, and break code — all without a single human checkpoint.