Mux’s new terms let it train its own AI models on customer data
Mux’s new terms let it train its own AI models on customer data

Stani Mihov
Founder & CEO
·

TL;DR
TL;DR
What changed:
Mux may now use customer data and output that do not contain personal data, or have been anonymized, “to train, fine-tune, or otherwise improve any AI Model owned or controlled by Mux.”
The old Terms of Service did not mention AI at all.
The new AI Services Addendum “applies to Customer’s use of the Services by default, without any separate activation or acceptance step by Customer.”
You now promise you have the consents needed for that AI training, and must keep records of end user consents for audio and video processing.
Mux will not defend you against claims that its AI output infringes someone else’s rights.
What to do: If you build on Mux, ask whether you can opt out of training, and check that your own terms cover AI processing of the video and audio you send.
The change
Mux provides “video infrastructure services to help development teams build video into their products,” such as video streaming through its API. On October 12, 2026, it updated its Terms of Service. The previous version was dated April 1, 2025.
The document is now called “Mux Terms of Service including AI Services Addendum.” The addendum covers Mux’s AI features, such as “transcription, summarization, content moderation, translation, chapter generation, and related capabilities.”
What changed
A right to train Mux’s own AI. Mux “may use Customer Data and Output, in each case that do not contain Personal Data or are or have been Anonymized, to train, fine-tune, or otherwise improve any AI Model owned or controlled by Mux.”
On by default. The addendum “applies to Customer’s use of the Services by default, without any separate activation or acceptance step by Customer.”
Your data goes to AI vendors Mux picks. For AI requests, Mux sends the relevant data, “which may include video, audio, and associated metadata,” to outside AI vendors, and “selects the AI Vendor(s) used for each AI Request in its discretion.”
Consents are on you. You promise you have “any consents required for AI-based processing and for the AI Model training described in Section 3.2.”
No cover for AI output. Mux “has no obligation to defend, indemnify, or hold Customer harmless against any claim to the extent arising from AI Output.”
A narrower performance promise. Mux no longer promises that the services will perform in line with its documentation.
What Mux can now train on
The old Terms gave Mux a license to use customer data to “maintain and improve our products and services.” They said nothing about AI or model training.
The new license keeps that wording and adds a condition: Mux may use customer data and output “to train, fine-tune, or otherwise improve AI Models” it owns or controls, “solely as permitted by, and subject to the conditions in, Section 3.2 of the AI Services Addendum.”
Section 3.2 sets one limit. The data must “not contain Personal Data” or must have been anonymized, which the Terms define as transformed so that it “can no longer be reasonably linked to any identified or identifiable natural person.” The section is not limited to data sent through the AI features. “Customer Data” covers “any data transmitted on or through the Services,” including “text, images, data, video, audio.”
The addendum also says Section 3.2 will “survive any termination or expiration of this AI Services Addendum.”
What it means for your users
Companies that build on Mux often send it video and audio recorded by their own users. The new terms put the legal groundwork for that on you.
For transcription and other audio features, you are “solely responsible for obtaining and maintaining all required consents” where the law requires them. You must “maintain appropriate records of all End User Consents” and provide evidence to Mux “upon Mux’s reasonable written request.”
The addendum also bans some uses outright. You may not use the AI features “for any High Risk Activity,” which includes using AI output as the basis for decisions on “employment, credit, housing, insurance, or law enforcement,” or for “clinical, diagnostic, or therapeutic purposes in healthcare.” You also may not use AI output to train your own models, except models Mux provides.
Other changes in the update
Documentation. The old performance warranty promised the services would perform “in material conformance with our Documentation and any Service Level Agreement.” The new one covers only “any Service Level Agreement agreed to between the parties.”
Delivery. You may not “place any third-party content delivery network, proxy, or cache in front of the Services,” unless the documentation allows it.
Free tiers. Free, trial, discounted and credit-based tiers “are provided without any service level commitment and may be modified, limited, or withdrawn at any time.”
AI changes. Mux “may modify, suspend, or discontinue any aspect of the AI Services at any time,” including the AI vendors and models it uses.
What improved
Mux “contractually requires each AI Vendor to refrain from using Customer Data or AI Output to train, fine-tune, or otherwise improve any AI Model.”
Mux will use “commercially reasonable efforts” to have AI vendors delete or return your data promptly after processing.
The Terms now say you own AI output as well as your data: “As between you and Mux, you own all Customer Data and Output.”
The current list of AI vendors is kept in Exhibit 3 of Mux’s Data Processing Addendum, and the Terms now link to an archive of previous versions of Mux’s legal documents.
Why this matters
Video and audio often contain faces, voices and names, and the training right depends on whether each piece of data counts as personal data or has been anonymized. Mux makes that call, and the Terms do not describe how.
More vendors are writing AI training into their contracts. Lovable now trains its AI on your prompts and code by default, Replit dropped its no-training promise for Pro customers, and TaxDome can now use client data to improve its own products. Mux’s line on AI output follows another pattern: Ramp says you own its AI output and carry the risk if it infringes.
Clauses like these arrive inside a routine update, which is why it pays to monitor vendor terms of service and keep track of the AI subprocessors your vendors use.
Potential impact
If you build on Mux, the update raises four practical questions:
Does your privacy notice cover AI processing and model training for the video and audio your users upload?
Can Mux exclude your account from training under a signed order? The addendum applies “except as otherwise expressly agreed in a Signed Order.”
Which AI vendors in Exhibit 3 of the DPA will receive your data?
Who reviews AI transcripts, summaries and moderation signals before they reach your users?
How Venpo detected it
Venpo monitors Mux’s legal pages and flagged the rewrite with every inserted and deleted line marked. The record is on the Mux Terms of Service change page on Venpo. Every quote in this article was checked against that redline.
Business outcome
Teams that catch this update can ask Mux about training before their next renewal and update their own user notices first. That is what continuous monitoring of vendor legal changes makes possible, and why it belongs in continuous vendor risk monitoring.
Key takeaway
Since October 12, 2026, Mux’s Terms of Service let it train its own AI models on customer data that contains no personal data or has been anonymized, apply that rule to every customer by default, and leave the consents for AI processing to you.
The change
Mux provides “video infrastructure services to help development teams build video into their products,” such as video streaming through its API. On October 12, 2026, it updated its Terms of Service. The previous version was dated April 1, 2025.
The document is now called “Mux Terms of Service including AI Services Addendum.” The addendum covers Mux’s AI features, such as “transcription, summarization, content moderation, translation, chapter generation, and related capabilities.”
What changed
A right to train Mux’s own AI. Mux “may use Customer Data and Output, in each case that do not contain Personal Data or are or have been Anonymized, to train, fine-tune, or otherwise improve any AI Model owned or controlled by Mux.”
On by default. The addendum “applies to Customer’s use of the Services by default, without any separate activation or acceptance step by Customer.”
Your data goes to AI vendors Mux picks. For AI requests, Mux sends the relevant data, “which may include video, audio, and associated metadata,” to outside AI vendors, and “selects the AI Vendor(s) used for each AI Request in its discretion.”
Consents are on you. You promise you have “any consents required for AI-based processing and for the AI Model training described in Section 3.2.”
No cover for AI output. Mux “has no obligation to defend, indemnify, or hold Customer harmless against any claim to the extent arising from AI Output.”
A narrower performance promise. Mux no longer promises that the services will perform in line with its documentation.
What Mux can now train on
The old Terms gave Mux a license to use customer data to “maintain and improve our products and services.” They said nothing about AI or model training.
The new license keeps that wording and adds a condition: Mux may use customer data and output “to train, fine-tune, or otherwise improve AI Models” it owns or controls, “solely as permitted by, and subject to the conditions in, Section 3.2 of the AI Services Addendum.”
Section 3.2 sets one limit. The data must “not contain Personal Data” or must have been anonymized, which the Terms define as transformed so that it “can no longer be reasonably linked to any identified or identifiable natural person.” The section is not limited to data sent through the AI features. “Customer Data” covers “any data transmitted on or through the Services,” including “text, images, data, video, audio.”
The addendum also says Section 3.2 will “survive any termination or expiration of this AI Services Addendum.”
What it means for your users
Companies that build on Mux often send it video and audio recorded by their own users. The new terms put the legal groundwork for that on you.
For transcription and other audio features, you are “solely responsible for obtaining and maintaining all required consents” where the law requires them. You must “maintain appropriate records of all End User Consents” and provide evidence to Mux “upon Mux’s reasonable written request.”
The addendum also bans some uses outright. You may not use the AI features “for any High Risk Activity,” which includes using AI output as the basis for decisions on “employment, credit, housing, insurance, or law enforcement,” or for “clinical, diagnostic, or therapeutic purposes in healthcare.” You also may not use AI output to train your own models, except models Mux provides.
Other changes in the update
Documentation. The old performance warranty promised the services would perform “in material conformance with our Documentation and any Service Level Agreement.” The new one covers only “any Service Level Agreement agreed to between the parties.”
Delivery. You may not “place any third-party content delivery network, proxy, or cache in front of the Services,” unless the documentation allows it.
Free tiers. Free, trial, discounted and credit-based tiers “are provided without any service level commitment and may be modified, limited, or withdrawn at any time.”
AI changes. Mux “may modify, suspend, or discontinue any aspect of the AI Services at any time,” including the AI vendors and models it uses.
What improved
Mux “contractually requires each AI Vendor to refrain from using Customer Data or AI Output to train, fine-tune, or otherwise improve any AI Model.”
Mux will use “commercially reasonable efforts” to have AI vendors delete or return your data promptly after processing.
The Terms now say you own AI output as well as your data: “As between you and Mux, you own all Customer Data and Output.”
The current list of AI vendors is kept in Exhibit 3 of Mux’s Data Processing Addendum, and the Terms now link to an archive of previous versions of Mux’s legal documents.
Why this matters
Video and audio often contain faces, voices and names, and the training right depends on whether each piece of data counts as personal data or has been anonymized. Mux makes that call, and the Terms do not describe how.
More vendors are writing AI training into their contracts. Lovable now trains its AI on your prompts and code by default, Replit dropped its no-training promise for Pro customers, and TaxDome can now use client data to improve its own products. Mux’s line on AI output follows another pattern: Ramp says you own its AI output and carry the risk if it infringes.
Clauses like these arrive inside a routine update, which is why it pays to monitor vendor terms of service and keep track of the AI subprocessors your vendors use.
Potential impact
If you build on Mux, the update raises four practical questions:
Does your privacy notice cover AI processing and model training for the video and audio your users upload?
Can Mux exclude your account from training under a signed order? The addendum applies “except as otherwise expressly agreed in a Signed Order.”
Which AI vendors in Exhibit 3 of the DPA will receive your data?
Who reviews AI transcripts, summaries and moderation signals before they reach your users?
How Venpo detected it
Venpo monitors Mux’s legal pages and flagged the rewrite with every inserted and deleted line marked. The record is on the Mux Terms of Service change page on Venpo. Every quote in this article was checked against that redline.
Business outcome
Teams that catch this update can ask Mux about training before their next renewal and update their own user notices first. That is what continuous monitoring of vendor legal changes makes possible, and why it belongs in continuous vendor risk monitoring.
Key takeaway
Since October 12, 2026, Mux’s Terms of Service let it train its own AI models on customer data that contains no personal data or has been anonymized, apply that rule to every customer by default, and leave the consents for AI processing to you.
Real-time change notifications
Stay ahead of every legal change
Get updates, product news and expert tips on navigating legal changes
Dispute resolution clause now requires mandatory arbitration in all regions
Data retention period extended from 2 years to 5 years for all services
New restrictions on AI-generated content in product descriptions
Third-party data sharing expanded to include analytics partners
Real-time change notifications
Stay ahead of every legal change
Get updates, product news and expert tips on navigating legal changes
Dispute resolution clause now requires mandatory arbitration in all regions
Data retention period extended from 2 years to 5 years for all services
New restrictions on AI-generated content in product descriptions
Third-party data sharing expanded to include analytics partners
