Amazon Employees Using ChatGPT for Coding and Customer Service Warned Not to Share Company Information With AI Chatbot
Amazon lawyers are warning employees against sharing company information with OpenAI’s ChatGPT, even as some at the tech giant are using the generative AI chatbot to assist them in answering customer questions, writing code, and other activities, according to a report from Business Insider. The lawyers cite uncertain ChatGPT data privacy policies as a reason not to share confidential information and code, despite a reported rise in productivity from some at Amazon using it.
Insider cited internal Amazon Slack channels full of frenzied discussion around the rules for using ChatGPT since it came out in November. The corporate lawyer who responded mentioned problems of data security involving the chatbot. They also mentioned that ChatGPT has responded in a way suggesting it had access to internal Amazon data. The potential for ChatGpt to use what Amazon employees write to train ChatGPT is a big reason for Amazon employees to avoid using ChatGPT for work purposes since the company “wouldn’t want its output to include or resemble our confidential information (and I’ve already seen instances where its output closely matches existing material),” the lawyer wrote.
OpenAI hasn’t laid out what data it uses to train ChatGPT or how user interactions inform that training. Amazon’s lawyer expressed the concern that corporate information the company doesn’t want to be shared might end up embedded in ChatGPT’s training. Amazon hasn’t banned access to ChatGPT, but employees who visit the website for the chatbot are warned that it is a third-party service not approved officially by Amazon Security. If employees want to use it anyway, they acknowledge the warning and open the chatbot.
Plenty of employees have already done so. ChatGPT has reportedly solved coding problems used for Amazon technical interviews and is aiding in other programming projects. AWS staff have also successfully employed ChatGPT to respond to customer questions. The AI chatbot has produced troubleshooting guides for database problems and helped write technical training manuals. Apparently, it was lackluster only at composing an epic rap battle.
The internal discussion also included a notable hint at how Amazon plans to respond to the explosion of interest in generative AI. The lawyer wrote that Amazon is working on comparable technology related to Alexa and the CodeWhisperer code recommendation product. Released around the same time as GitHub released its GPT-3-descended Copilot programming assistant, CodeWhisperer learns from the user’s coding and comments, adapting to individual style and context over time. Unlike Copilot, CodeWhisperer defaults to producing new code that is not in its database where possible. It will highlight any suggestions it makes that may be too close to programming owned by someone else, and indicate who has the license for that code.