Amazon’s cloud division reportedly suffered outages last year involving its own AI systems.
A December disruption lasting about 13 hours was blamed on an AI agent that deleted and rebuilt part of its environment, according to reports.
Only one incident is said to have affected customer-facing services.
Amazon said the problem resulted from user error and misconfigured access controls, not from artificial intelligence itself.
The company added that the interruption was limited and did not affect core services such as compute, storage or databases.
It also introduced extra safeguards, including mandatory peer review for production access.
The incidents have intensified debate about the growing use of AI inside critical infrastructure.
Security experts argue that autonomous tools can act without fully understanding wider consequences.
They say humans usually have more time to notice mistakes when performing the same tasks manually.
The issue comes as Andy Jassy pushes efficiency changes and job cuts across Amazon.
The company announced thousands of layoffs but insists they are not directly replacing staff with AI.
AWS underpins large parts of the internet and holds major public-sector contracts.
Previous outages have already raised concerns about the concentration of online services in a few providers.
