Amazon’s new AI chatbot, Amazon Q, has reportedly been leaking internal information in a series of severe hallucinations.
Internal documents, which were obtained by Platformer, reveal that the chatbot has inadvertently disclosed confidential information.
This includes the locations of Amazon Web Services (AWS) data centers, details about internal discount programs, and features not yet released to the public.
The situation was allegedly deemed serious enough to be classified as “sev 2,” prompting engineers to work extended hours, including nights and weekends, to address the issue.
In response to speculation, an Amazon spokesperson said that no security breaches had been detected due to this incident. They emphasized the company’s commitment to continually refining Amazon Q, particularly as it progresses from a preview product to general availability.
Contradicting the internal documents, a spokesperson for DCD stated that Amazon Q has not leaked any confidential information.
Amazon Q is Amazon’s new business-centric chatbot, which bills sophisticated security and data sovereignty for business users among its headline features.
However, the internal documents indicate that Amazon Q has provided misleading information on various topics, causing discontent among early adopters and concerns among Amazon employees, as evidenced by comments shared in internal Slack channels.
One employee expressed concern over the chatbot’s responses, which could potentially expose Amazon to legal liabilities. None of this has been officially confirmed.
Amazon is known for maintaining strict confidentiality about its data center locations.
In 2018, WikiLeaks released internal Amazon documents outlining the locations of its data centers as of 2015. Since then, the exact locations of all Amazon data centers have remained undisclosed, except for instances where they have been revealed through planning document reports.
This is slightly reminiscent of Google demonstrating Bard for the first time in public only for it to make a serious factual error about the James Webb telescope, which the AI said “took the very first pictures of a planet outside of our own solar system.” This was incorrect.
AI hallucinations and bloopers are one thing, but leaking sensitive information is another, however.
