Zoom updates its legalese explicitly promising not to feed vidchats to AIs
Welcome – but weird – after months-old policy change that seemingly allowed it went viral
Vidchat and collaboration outfit Zoom has insisted it never intended to give the impression vidchats it hosts would be fed into an artificial intelligence model, after a delayed viral outrage to a March change to its legalese.
On March 31 2023, Zoom updated its terms and conditions, with changes including Clause 10.4 titled "Customer License Grant" that gave the vidchat conduit "a perpetual, worldwide, non-exclusive, royalty-free, sublicensable, and transferable license." The clause arguably gave Zoom the right to do a laundry list of stuff to "customer content" – including applying it to machine learning and artificial intelligence.
Zoom to mandate client updates every ninety days, starting Nov 1READ MORE
The changes went largely unnoticed at the time.
Zoom has since tweaked its legalese a couple more times. And readers of the update made on July 27 appear to have read clause 10.4, assumed it was entirely new, and balked at the prospect that their vidchats could be fed into the maw of Zoom's AI.
"Each uttered word, each shared screen, every slide presented – the canvas of our interactions lays bare, a sacrificial offering to the algorithms”, protested a LinkedIn post typical of many such missives that crossed Reg hacks' desks yesterday.
That disgruntled but delayed sentiment has delivered: on Monday Zoom updated its Terms of Service with the following addendum to clause 10.4:
Notwithstanding the above, Zoom will not use audio, video or chat Customer Content to train our artificial intelligence models without your consent.
- Boffins say they can turn typing sounds into text with 95% accuracy
- Microsoft puts profanity filter on %@!#ing Teams transcripts
- Snowflake and Zoom celebrate shrunken cloud bills as margins swell
- When you try to hire a freelancer to write SQL and all you get is incorrect AI garbage
Zoom's chief product officer Smita Hashim blogged about the change.
"It's important to us at Zoom to empower our customers with innovative and secure communication solutions," she wrote. "We've updated our terms of service (in section 10.4) to further confirm that we will not use audio, video, or chat customer content to train our artificial intelligence models without your consent."
Hashim explained that Zoom never intended to create the impression it would suck user's vidchats into AI for any purpose other than running and refining its own services. She argued that those who felt that might not be the case had misunderstood clause 10.4 and perhaps not considered the impact of other clauses that collectively spell out that Zoom doesn't own customers' content but has permission to provide value-added services based on their stuff.
Her post goes on to explain, and illustrate, Zoom's opt-in user interface. She also explains that Zoom uses generative AI for good – for instance with tools that summarize meetings if users request it.
"And even if you choose to share your data, it will not be used for training of any third-party models," Hashim added.
Several AI outfits have ruthlessly dredged the internet to build their models – almost always without a thought to paying creators. Fears have therefore understandably arisen that data shared with model-makers could be used to feed other AIs without user consent, with the potential for proprietary information to leak.
Given Zoom is used for many private communications, the prospect of a video session reaching the wider user population is terrifying.
As is the fact that sufficiently angry expressions of such sentiments took four months to surface, and that Zoom's legalese was that fuzzy for the same period. ®