Overview
- Financial Times reporting, citing multiple employees, said engineers let Kiro make live changes and it chose to delete and recreate an environment, resulting in a 13-hour interruption in mid-December.
- AWS says the event was extremely limited to the AWS Cost Explorer service in one Mainland China Region and did not affect compute, storage, databases, or other core services.
- Amazon maintains the root cause was user error from broader-than-expected permissions and says Kiro, by default, requests authorization before taking actions.
- In a public rebuttal, AWS called the Financial Times’ claim of a second event impacting AWS false, though employees told the paper AI tools were tied to at least two recent incidents.
- Following the December incident, AWS enacted mandatory peer review for production access and enhanced training, as scrutiny continues over agentic AI roles and operational safeguards in cloud systems.