Zoom Adjusts Policy on Generative AI Training With Customer Data Following Online Outcry
Zoom has hastily revised its terms of service (TOS) regarding generative AI and how it uses customer data following loud protests. The videoconferencing platform’s policy appeared to allow it to use customer data to train the large language models powering its generative AI features. A StackDiary post on Sunday called attention to the phrasing, sparking angry posts from many users, including prominent members of the tech and entertainment industries. Zoom denied this interpretation, but the company did end up “clarifying” the matter and rewriting the TOS anyway over the course of a day.
Zoom has rolled out a few generative AI features and even signed a deal to embed Anthropic’s Claude model while funding the startup. Those features led to an updated TOS and the current furor. The contentious terms were part of section 10.4 of Zoom’s TOS. The legal boilerplate seemed to suggest that Zoom can use customer data for training its AI without any ability to opt out of the program. The text, with the key terms highlighted, originally read:
“You agree to grant and hereby grant Zoom a perpetual, worldwide, non-exclusive, royalty-free, sublicensable, and transferable license and all other rights required or necessary to redistribute, publish, import, access, use, store, transmit, review, disclose, preserve, extract, modify, reproduce, share, use, display, copy, distribute, translate, transcribe, create derivative works, and process Customer Content and to perform all acts with respect to the Customer Content…for the purpose of product and service development, marketing, analytics, quality assurance, machine learning, artificial intelligence, training, testing, improvement of the Services, Software, or Zoom’s other products, services, and software, or any combination thereof.”
Though the terms were introduced in March, it was only after the new post started going viral on X (formerly Twitter) and elsewhere that Zoom felt it had to edit its phrasing. Angry comments appeared, written by the likes of actor and SAG-AFTRA AI advisor Justine Bateman and film and TV writer John Rogers, not to mention many non-famous Zoom users. Within 24 hours, Zoom had a blog post explaining it was a misunderstanding.
“When you choose to enable Zoom IQ Meeting Summary or Zoom IQ Team Chat Compose, you will also be presented with a transparent consent process for training our AI models using your customer content,” Zoom product chief Smita Hashim wrote in a blog post. “Your content is used solely to improve the performance and accuracy of these AI services. And even if you chose to share your data, it will not be used for training of any third-party models. “
New Terms, Same Service?
Hashim’s blog post corresponded to a change in Zoom’s TOS to take out the bits in section 10.4 about machine learning and AI. Instead, the clause has more boilerplate centered on Zoom’s rights to “Service Generated Data” and “Aggregated Anonymous Data.” The section ends with a separate sentence in bold font stating, “Notwithstanding the above, Zoom will not use audio, video or chat Customer Content to train our artificial intelligence models without your consent.”
Zoom doesn’t explain what Service Generated Data is, and Section 10.2 has its own reference to using it to train AI, again without an opt-out option. It’s an easy bet that other companies experimenting with generative AI are keeping a close eye on Zoom and how the public reacts, potentially setting a precedent that could have a much larger impact on the future of the industry.