Thursday, February 22, 2024
27 C
Brunei Town

Zoom says isn’t training AI on calls without consent, other data is fair game

AP – An update to Zoom’s terms of service is raising alarm bells on social media, with users claiming it revealed the videoconferencing company is now tapping their online doctor visits and virtual happy hours to train artificial intelligence (AI) models.

“Zoom terms of service now require you to allow AI to train on all your data – audio, facial recognition, private conversations – unconditionally and irrevocably, with no opt out,” read one widely-shared tweet this week that has since been deleted.

The company quickly responded with a blog post on Monday stressing that it “will not use audio, video, or chat customer content to train our artificial intelligence models without your consent”, and adding a line to the terms to make this clearer.

Online privacy experts said that this policy is now accurately reflected in the document.

However, the terms do still allow Zoom to train AI on other data, such as how customers behave – and they question how much choice some meeting participants will have to opt out if, say, their boss decides otherwise.

Here’s a look at the facts.

A display for Zoom Video Communications in New York, the United States. PHOTO: AP

CLAIM: Zoom’s terms of service give the company permission to use all customer data, including private conversations, for training AI, with no ability to opt out.

THE FACTS: That is not accurate now that Zoom has added language to the terms to make its policy clearer, experts said.

The current terms would not allow the company to tap user-generated content like video and chat for AI training without a customer opting in. However, once a meeting host agrees, other participants would have to leave if they don’t want to consent.

The terms also allow Zoom to use other data, including information about user behaviour, without additional permission. “The face of these terms of service does now assure the user that Zoom is not going to use their customer content for the purpose of training AI models without their consent,” Electronic Privacy Information Center Director of Litigation and Senior Counsel John Davisson told the Associated Press (AP).

At issue is language Zoom added to its terms in March. The document differentiates between two types of data: “service generated data”, such as what features customers use and what part of the world they are in, and “customer content”, which is the data created by users themselves, such as audio or chat transcripts.

The terms state that service-generated data can be used for “machine learning or AI (including the purposes of training and tuning algorithms and models”.

Zoom’s blog post said the company considers such data “to be our data”, and experts confirm this language would allow the company to use the data for AI training without obtaining additional consent. Separately, the terms said that customer content may also be used “for the purpose” of machine learning or AI.

The company clarified in its post that this refers to new generative AI features that users must agree to, which create things like automated meeting summaries for customers.

Zoom said in a statement to the AP that in addition to enabling the features for themselves, users must separately consent to sharing this data with the company.

Experts said that the language in the March update was wide-reaching and could have opened the door for the company to use that data without additional permission if it wanted to.

But Zoom added a more explicit caveat on Monday: “Notwithstanding the above, Zoom will not use audio, video or chat customer content to train our AI models without your consent.”

With this language, Davisson said that using such data to train AI without a user consenting would now constitute a violation of the terms on Zoom’s part, opening the company up to litigation.

However, experts said this could still pose problems for some participants in Zoom calls if the host opts into the generative AI features. Zoom said that if a meeting organiser uses the meeting summary feature, participants are sent a notification alerting them that an AI feature has been enabled and that their data may be shared for AI training. They are prompted to either proceed with the meeting or to leave.

Although this in theory offers the ability to control how their data is used, it may not be possible for someone to opt out of a meeting if they disagree, said Electronic Frontier Foundation Director of Policy and Advocacy Katharine Trendacosta.

“If the administrator consents and it’s your boss at your work who requires you to use Zoom, how is that really consent?” she asked.