Amazon’s cloud division suffered at least one outage last year after an internal AI agent made a critical change to its own environment.
The system reportedly deleted and recreated part of its setup, causing a 13-hour disruption.
AWS runs large parts of the internet, so even smaller incidents raise concern about concentration of infrastructure.
Another outage in October knocked dozens of websites offline for hours.
Amazon said the event resulted from user misconfiguration, not artificial intelligence failure.
It added that only one incident affected customer-facing services and that safeguards have since increased.
These include stricter access controls and mandatory peer review for sensitive actions.
The issue comes as Andy Jassy pushes efficiency through AI while cutting thousands of jobs.
Amazon insists the layoffs are about restructuring rather than replacing staff with automation.
Security specialists dispute the company’s explanation.
They argue AI agents act faster than humans and may lack full operational context.
That can increase the risk of unintended system changes.
Experts say complex AI systems will always carry some unpredictability.
The incidents have intensified debate over how far companies should automate critical infrastructure.
