You really shouldn’t give AI free rein to do anything it wants on your system.
Amazon Web Services (AWS) has reportedly suffered from a couple of outages due to misbehaving AI agents. According to the Financial Times, the most recent interruption happened in December last year, when its Koiro AI coding tool decided to erase the environment it was working on, resulting in a 13-hour disruption.
“We’ve already seen at least two production outages,” one senior AWS employee told the publication. “The engineers let the AI resolve an issue without intervention. The outages were small but entirely foreseeable.”
On the other hand, the company reported that the incidents were relatively minor, with the December disruption only affecting a single service in parts of mainland China, and the other one having no effect on customer-facing services. Amazon said to the Financial Times that the involvement of AI tools in these incidents was just a coincidence, and that the same issue would have occurred even with other developer tools or through manual action. “In both instances, this was user error, not AI error,” the company said.
Correcting the Financial Times report about AWS, Kiro, and AI
We want to address the inaccuracies in the Financial Times’ reporting yesterday. The brief service interruption they reported on was the result of user error—specifically misconfigured access controls—not AI as the story claims.
https://www.aboutamazon.com/news/aws/aws-service-outage-ai-bot-kiro