After backlash, Zoom reversed its terms of service so it will not use customer content to train AI (August 2023)

The problem

In March 2023 Zoom updated its online terms of service, with the change taking effect 27 July 2023. Section 10.4 granted Zoom broad rights over “Service Generated Data for any purpose,” and, as source quoted the terms directly, the permitted purposes included “machine learning or artificial intelligence (including for the purposes of training and tuning of algorithms and models)” (source). There was no opt-out. A second, independent newsroom quoted a different part of the same clause: source reported that “In Section 10.4 of Zoom“‘s terms, “updated in March,” users granted Zoom a licence for purposes including “machine learning, artificial intelligence, training, testing” (source). On 6–7 August 2023 the clause was surfaced publicly and a backlash spread among users, educators and privacy advocates who read it as permission to feed calls, chats and other communications into AI training. Zoom’s own blog later confirmed the origin firsthand: “We changed our online terms of service in March 2023” (source).

What happened

After the backlash, Zoom reversed course and updated its terms. In its first-party source (attributed to Chief Product Officer Smita Hashim, “edited on August 11, 2023”), the company said it “does not use any of your audio, video, chat, screen sharing, attachments, or other communications like customer content … to train Zoom’s or third-party artificial intelligence models,” and that it had moved to “reorganize Section 10” of the terms (source). source reported the company had updated the terms so that “we will not use audio, video, or chat Customer Content to train our artificial intelligence models without customer consent” (source). source independently reported the broadened statement, that Zoom “does not use any of your audio, video, chat, screen-sharing, attachments, or other communications like customer content” to train AI, noting the change to “the previously troublesome section 10 in” Zoom’s terms of service (source).

The distinctive fact

The reversal carried a consent caveat rather than a blanket prohibition. As source reported the clarification, Zoom said that “for AI, we do not use audio, video or chat content for training our models without customer consent”, and users who switch on Zoom’s generative-AI features (Meeting Summary, Team Chat Compose) consent to Zoom training on their content (source). The walk-back removed the automatic, no-opt-out AI-training grant over communications; it did not remove AI training entirely.

The weakest link (source honesty)

The single soft spot is that the original, pre-walk-back section 10.4 text is quoted here through independent press (CNBC and NBC News), not from an archived first-party copy of Zoom’s old terms page. CNBC and NBC News are strong independent secondaries (Tier 2), and they quote the clause verbatim and agree on its substance, but the canonical original terms document itself is not in hand. Zoom’s own blog confirms first-party only that the terms changed in March 2023, not the exact wording of the disputed clause. Retrieving and archiving Zoom’s own pre-walk-back section 10.4 (or a recorded Zoom confirmation of the before/after text) is the outstanding step this page leaves open.

How this was verified

• Method: AI Terms-of-Service walk-back record. Both critical claims, the original AI/ML-training grant (c1) and the post-backlash reversal (c2), rest on two independent newsrooms each, plus Zoom’s own first-party blog as the origin. Every load-bearing fragment was re-checked verbatim against the specific capture it is attributed to (12/12 bind, 0 misses), this session.
• Byte-tie: all five on-disk captures are Wayback id_ raw payloads whose local sha1-b32 equals the live Wayback CDX digest (status 200): Zoom blog 20230812065407, CNBC 20230807175926, NBC News 20230808022407, TechCrunch 20230809013803, Engadget 20230811184916.
• Date checked: 2026-08-09 (graduation); presentation re-verified 2026-08-18.
• Ceiling: amber. Green is a human-only step; it would require a recorded confirmation from Zoom of the exact date section 10.4 was revised and the canonical before/after clause text, plus an archived first-party copy of the pre-walk-back terms page. This badge is not verified.

Sources

1. Zoom blog · “How Zoom’s terms of service and practices apply to AI” (Smita Hashim, Chief Product Officer) · edited 2023-08-11 · Tier 1 (first-party origin) · https://blog.zoom.us/zooms-term-service-ai/
2. CNBC · “Zoom AI tools trained using some customer data, updated terms say” · 2023-08-07 · Tier 2 (independent secondary) · https://www.cnbc.com/2023/08/07/zoom-ai-tools-trained-using-some-customer-data.html
3. NBC News · “Zoom responds to privacy concerns raised by AI data collection” · 2023-08-07 · Tier 2 (independent secondary) · https://www.nbcnews.com/tech/innovation/zoom-ai-privacy-tos-terms-of-service-data-rcna98665
4. TechCrunch · “Zoom knots itself a legal tangle over use of customer data for training AI models” (Natasha Lomas) · 2023-08-08 · Tier 2 (independent secondary) · https://techcrunch.com/2023/08/08/zoom-data-mining-for-ai-terms-gdpr-eprivacy/
5. Engadget · “Zoom now says it won’t use any customer content for AI training” (Will Shanklin) · 2023-08-11 · Tier 2 (independent secondary) · https://www.engadget.com/zoom-now-says-it-wont-use-any-customer-content-for-ai-training-170533411.html

Related case files

• Adobe adds a “we don’t train AI on your content” clarification after backlash (2024): the same ToS/AI-training walk-back pattern, one year later.
• Dropbox adds a third-party-AI toggle and confirms customer data is not used to train models (2023): a platform granting the opt-out Zoom’s users demanded.
• WeTransfer removes an AI-training clause from its terms after backlash (2025): a near-identical terms reversal in the file-transfer category.
• Slack updates its AI privacy principles after user backlash over ML training opt-out (2024): another communications tool clarifying customer-content training.