When Amazon announced its business-focused AI chatbot Q, the company framed it as a secure assistant that could help employees query internal documents, write code, and summarize reports. Within days, internal complaints leaked to the press claiming that Q was hallucinating badly and even revealing confidential information. That combination led to viral posts warning that Amazon’s “AI” was already out of control.

What Q is supposed to do

Amazon Q is marketed as an AI assistant for companies using Amazon Web Services. It can be customized to answer questions about a company’s own documents and systems, generate code with the help of Amazon CodeWhisperer, and power customer support tools.

Unlike consumer chatbots that pull information largely from the public web, Q is designed to work primarily with private corporate data. That is part of its appeal and also the root of the privacy concerns.

The leaked complaints

In late 2023, tech outlets reported on internal Amazon documents describing “severe hallucinations” and alleged leaks of confidential info during Q’s public preview. Examples included the bot supposedly giving out details about internal discounts, unreleased features, and even data center locations.

The incident was reportedly tagged as a serious internal issue, and employees warned that the behavior could create real risk if customers relied on Q for sensitive work.

Amazon’s response

Amazon has pushed back on the most alarming interpretations. Company spokespeople say that feedback from employees is part of the normal testing process and that no verified security breach occurred. Q, they argue, uses multiple models and access controls designed to keep customer data compartmentalized, and the company continues to refine the system as it moves from preview toward general availability.

At the same time, Amazon has acknowledged that, like any large language model, Q can generate incorrect answers – the “hallucinations” that are a known limitation of this type of AI.

Rumor vs reality

Online discussion often compresses these nuances into a simple narrative that Amazon’s AI “leaks secrets” or is “not really secure.” The documents that leaked show that employees were worried about potential exposure and inaccurate answers, but public reporting so far has not confirmed a case where Q exfiltrated private customer data into the wild.

What is real is that early versions of Q appear to have surfaced information that employees did not expect it to share inside Amazon’s own environment, and that alone is enough to make outside customers cautious. It is also clear that Amazon is racing competitors in the enterprise AI space and is under pressure to ship quickly, which raises legitimate questions about how thoroughly any of these systems are being tested before launch.

What users should keep in mind

• Q is a real product from Amazon that is still evolving, not a fictional super-intelligence, and it has the same limitations and risks as other large language models.
• Reports of “severe hallucinations” and internal concern are based on leaked documents, while Amazon disputes that any confirmed security breach has taken place.
• Companies considering Q or similar tools should treat them as experimental and avoid feeding highly sensitive data into any AI system without strong contractual and technical safeguards.

The story around Amazon Q shows how quickly early technical problems can turn into sweeping online rumors. The truth so far sits in the middle: Q is neither a magical all-knowing assistant nor a cartoon villain that has already dumped corporate secrets onto the open internet, but a work-in-progress AI tool that deserves close scrutiny.