Amazon’s cloud division suffered outages last year that were reportedly triggered by its own AI systems.
One disruption in December lasted about 13 hours after an AI agent deleted and rebuilt part of its environment.
Another earlier incident caused shorter service problems.
AWS said only one event affected customer-facing services.
The company blamed misconfigured access controls and called the failures “user error”, not AI error.
It added that new safeguards and mandatory peer reviews are now in place.
The outages come as Amazon cuts thousands of jobs while expanding AI use.
Chief executive Andy Jassy has said AI will improve efficiency but not replace staff directly.
Some cybersecurity experts dispute Amazon’s explanation.
They argue AI agents can act quickly without fully understanding wider consequences.
Unlike human engineers, they may execute harmful actions before anyone intervenes.
The incidents have renewed concern about heavy reliance on a few cloud providers.
They also highlight the risks of deploying autonomous AI in critical infrastructure.
