Top insights for IT pros
From cybersecurity and big data to cloud computing, IT Brew covers the latest trends shaping business tech in our 4x weekly newsletter, virtual events with industry experts, and digital guides.
Amazon’s experimental generative AI is…not doing so hot, according to a recent report by Platformer.
The company announced its Q chatbot during Amazon Web Service’s (AWS) annual developer conference—and it’s already “experiencing severe hallucinations and leaking confidential data,” the site reported. The confidential data in question includes the location of AWS data centers, internal discount programs, and unreleased features, according to leaked documents obtained by Platformer.
One of the incidents was marked “sev 2,” referring to a bug so severe that Amazon contacted engineers during the night and made them work through the weekend to solve it, according to Platformer. While companies define their own incident severity levels, a ranking of 2 generally refers to an incident with a major impact.
Amazon is hoping to compete with similar AI offerings by Google and Microsoft-tied OpenAI, and recently announced a $4 billion investment in AI startup Anthropic. Those competitors have had their own issues with hallucinating AIs: Microsoft’s ChatGPT-supported Bing chatbot infamously went off the rails soon after release, while Google’s Bard served up incorrect information during its very first public demo. Yet Amazon’s pitch for Q included the claim that its approach would minimize the risk of losing control of sensitive internal data.
“We think Q has the potential to become a work companion for millions and millions of people in their work life,” Adam Selipsky, CEO of AWS, told the New York Times. Selipsky added that Amazon had designed Q in response to reports other companies had “banned these AI assistants from the enterprise because of the security and privacy concerns,” and Q has safeguards in place such as need-to-know restrictions on corporate data.
In a statement to IT Brew, Amazon said reports of Q misbehaving derived from the normal feedback process and that the company had not determined any actual security issues.
“Some employees are sharing feedback through internal channels and ticketing systems, which is standard practice at Amazon,” Amazon spokesperson Ryan C. Peters wrote. “No security issue was identified as a result of that feedback. We appreciate all of the feedback we’ve already received and will continue to tune Q as it transitions from being a product in preview to being generally available.”
“Amazon Q has not leaked confidential information,” the statement continued.
Q relies on an Amazon platform called Bedrock, and it links several generative AI models including Meta’s and Anthropic’s, as well as Titan, its in-house AI image generator, according to the Times. Amazon has set initial pricing at $20 per user per month.
While Amazon is less far along the AI curve than OpenAI or Google, the Wall Street Journal recently reported that AWS’s revenue has remained healthy and that it’s developing AI chips to compete with industry leader Nvidia.