An empirical review of the Terms of Service of Meta, TikTok, Snap, YouTube, and Google — examining how each platform disclaims liability for the content its algorithms select, rank, and deliver to users.
By Andrew Leahey · Published February 2026 · 16 clauses analyzed · 7 documents reviewed
Litigation Context
Hundreds of lawsuits filed in federal and state courts — including the consolidated multidistrict litigation In re Social Media Adolescent Addiction/Personal Injury Products Liability Litigation (MDL No. 3047, N.D. Cal.) — allege that social media platforms designed their recommendation algorithms to be habit-forming and addictive, particularly for teenagers and children, causing mental health injuries including anxiety, depression, eating disorders, and self-harm.
The core product-liability theory: the algorithm itself is the product, and its design — optimizing for engagement over user wellbeing — makes it defective. Plaintiffs argue this is distinct from the third-party content delivered, which is shielded by Section 230 of the Communications Decency Act.
This case study examines the contractual language these companies have put in place to disclaim exactly this kind of liability. A central finding: every platform contractually defines what it provides as a unitary "Service" — and the recommendation algorithm is part of that Service, not carved out as third-party content or user speech. The liability limitations, damage exclusions, and indemnification requirements that apply to server uptime and login screens apply equally to the algorithm that decides what a thirteen-year-old sees next.
1. Platform Coverage
We reviewed the current Terms of Service for the six platforms most frequently named in the algorithmic addiction litigation. The table below shows our data coverage and which liability-related clauses we extracted from each.
| Platform | Versions | Earliest | Liability Cap | Indemnification | Arbitration | Class Waiver |
|---|---|---|---|---|---|---|
| Meta Platforms | ||||||
| Meta — Terms of Service | 2 | Nov 2005 | ✓ | ✓ | — | — |
| Facebook — Terms of Service | 5 | Jan 2026 | ✓ | — | — | — |
| WhatsApp — Terms of Service | 6 | Dec 2020 | ✓ | ✓ | ✓ | — |
| Alphabet / Google | ||||||
| Google — Terms of Service | 1 | Jan 2026 | ✓ | ✓ | — | — |
| YouTube — Terms of Service | 1 | Jan 2026 | ✓ | ✓ | — | — |
| Snap Inc. | ||||||
| Snap — Terms of Service | 5 | Jan 2017 | ✓ | ✓ | ✓ | — |
| TikTok / ByteDance | ||||||
| TikTok — Terms of Service | 3 | Mar 2021 | ✓ | ✓ | ✓ | — |
Compare all Tech liability clauses → Compare all Tech indemnification clauses → Compare all Tech arbitration clauses →
2. Liability Limitations and Damage Caps
Every platform caps its liability exposure through broad limitation-of-liability clauses. These clauses exclude consequential, indirect, incidental, and punitive damages — the exact categories of harm alleged in the algorithmic addiction litigation (emotional distress, mental health injuries, loss of wellbeing). Google and TikTok go further with explicit "as is" disclaimers — Google states it provides services "AS IS" WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING... ACCURACY, RELIABILITY, AVAILABILITY, OR ABILITY TO MEET YOUR NEEDS," and TikTok provides its platform "as is" with no guarantees that it "will be safe, secure, and free from errors." The remaining platforms achieve the same practical result through blanket damage exclusions without using the specific words "as is." Because each platform defines the recommendation algorithm as part of its unitary "Service" (see Section 5), these damage caps cover the algorithmic recommendation system by inclusion — no platform names the algorithm specifically, and none carves it out.
Meta Platforms — Meta Terms of Service
EXCEPT IN JURISDICTIONS WHERE SUCH PROVISIONS ARE RESTRICTED, IN NO EVENT WILL FACEBOOK BE LIABLE TO YOU OR ANY THIRD PERSON FOR ANY INDIRECT, CONSEQUENTIAL, EXEMPLARY, INCIDENTAL, SPECIAL OR PUNITIVE DAMAGES, INCLUDING ALSO LOST PROFITS ARISING FROM YOUR USE OF THE WEB SITE OR THE SERVICE, EVEN IF FACEBOOK HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. NOTWITHSTANDING ANYTHING TO THE CONTRARY CONTAINED HEREIN, FACEBOOK'S LIABILITY TO YOU FOR ANY CAUSE WHATSOEVER, AND REGARDLESS OF THE FORM OF THE ACTION, WILL AT ALL TIMES BE LIMITED TO THE AMOUNT PAID, IF ANY, BY YOU TO FACEBOOK FOR THE SERVICE DURING THE TERM OF MEMBERSHIP.
Meta (Facebook) — Version 1271, retrieved November 26, 2005
Snap Inc. — Snapchat Terms of Service
Snap — Version 844727, retrieved October 13, 201716. Limitation of Liability
TO THE MAXIMUM EXTENT PERMITTED BY LAW, SNAP INC. AND OUR MANAGING MEMBERS, SHAREHOLDERS, EMPLOYEES, AFFILIATES, LICENSORS, AGENTS, AND SUPPLIERS WILL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, PUNITIVE, OR MULTIPLE DAMAGES, OR ANY LOSS OF PROFITS OR REVENUES, WHETHER INCURRED DIRECTLY OR INDIRECTLY, OR ANY LOSS OF DATA, USE, GOODWILL, OR OTHER INTANGIBLE LOSSES, RESULTING FROM: (A) YOUR ACCESS TO OR USE OF OR INABILITY TO ACCESS OR USE THE SERVICES; (B) THE CONDUCT OR CONTENT OF OTHER USERS OR THIRD PARTIES ON OR THROUGH THE SERVICES; OR (C) UNAUTHORIZED ACCESS, USE, OR ALTERATION OF YOUR CONTENT, EVEN IF SNAP INC. HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. IN NO EVENT WILL SNAP INC.’S AGGREGATE LIABILITY FOR ALL CLAIMS RELATING TO THE SERVICES EXCEED THE GREATER OF $100 USD OR THE AMOUNT YOU PAID SNAP INC., IF ANY, IN THE LAST 12 MONTHS.
TikTok / ByteDance — TikTok Terms of Service
TikTok — Version 2265, retrieved March 30, 202110. LIMITATION OF LIABILITY
NOTHING IN THESE TERMS SHALL EXCLUDE OR LIMIT OUR LIABILITY FOR LOSSES WHICH MAY NOT BE LAWFULLY EXCLUDED OR LIMITED BY APPLICABLE LAW. THIS INCLUDES LIABILITY FOR DEATH OR PERSONAL INJURY CAUSED BY OUR NEGLIGENCE OR THE NEGLIGENCE OF OUR EMPLOYEES, AGENTS OR SUBCONTRACTORS AND FOR FRAUD OR FRAUDULENT MISREPRESENTATION.
Alphabet / Google — YouTube Terms of Service
YouTube — Version 73, retrieved January 31, 2026Limitation of Liability
EXCEPT AS REQUIRED BY APPLICABLE LAW, YOUTUBE, ITS AFFILIATES, OFFICERS, DIRECTORS, EMPLOYEES AND AGENTS WILL NOT BE RESPONSIBLE FOR ANY LOSS OF PROFITS, REVENUES, BUSINESS OPPORTUNITIES, GOODWILL, OR ANTICIPATED SAVINGS; LOSS OR CORRUPTION OF DATA; INDIRECT OR CONSEQUENTIAL LOSS; PUNITIVE DAMAGES CAUSED BY:
Alphabet / Google — Google Terms of Service
Alphabet (Google) — Version 3, retrieved January 31, 2026Warranty disclaimer
TO THE EXTENT ALLOWED BY APPLICABLE LAW, WE PROVIDE OUR SERVICES "AS IS" WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, AND NON-INFRINGEMENT. FOR EXAMPLE, WE DON'T MAKE ANY WARRANTIES ABOUT THE CONTENT OR FEATURES OF THE SERVICES, INCLUDING THEIR ACCURACY, RELIABILITY, AVAILABILITY, OR ABILITY TO MEET YOUR NEEDS.
Liabilities
These terms only limit our responsibilities as allowed by applicable law. These terms don't limit liability for gross negligence or willful misconduct.
To the extent allowed by applicable law:
Google is liable only for its breaches of these terms or applicable service-specific additional terms.
Google isn't liable for: loss of profits, revenues, business opportunities, goodwill, or anticipated savings; indirect or consequential losses; punitive damages.
Google's total liability arising out of or relating to these terms is limited to the greater of (1) $200 or (2) the fees paid to use the relevant services in the 12 months before the dispute.
Finding 1
Every platform excludes liability for consequential, indirect, and punitive damages — the exact categories of harm alleged in the algorithmic addiction litigation. Google and TikTok explicitly provide their services "as is" with no express or implied warranties; the others achieve the same practical result through blanket damage exclusions. None carves out algorithmic content recommendation, personalization, or the "For You" feed from these limitations. The algorithm is covered by the same liability cap as any other part of the "Service."
3. Indemnification: Users Bear the Risk
Beyond disclaiming their own liability, platforms require users to indemnify the company — meaning users agree to cover the platform's legal costs if the user's activity leads to claims. In the context of algorithmic harm litigation, indemnification clauses are significant because they attempt to shift responsibility for any harm associated with service use back to the user.
Snap Inc. — Snapchat Terms of Service
Snap's indemnification clause is particularly notable because it explicitly covers content or services "even if recommended, made available, or approved by Snap":
You agree, to the extent permitted by law, to indemnify, defend, and hold harmless Snap Inc., our affiliates, directors, officers, stockholders, employees, licensors, and agents from and against any and all complaints, charges, claims, damages, losses, costs, liabilities, and expenses (including attorneys’ fees) due to, arising out of, or relating in any way to: (a) your access to or use of the Services; (b) your content; and (c) your breach of these Terms.
Snap — Version 844727, retrieved October 13, 2017
Finding 2
Snap's indemnification clause explicitly addresses the recommendation scenario — users must indemnify Snap even for products or services that Snap recommended. This directly anticipates claims that Snap's recommendation algorithm caused harm: the user contractually agrees to hold Snap harmless regardless.
Alphabet / Google — YouTube
To the extent permitted by applicable law, you agree to defend, indemnify and hold harmless YouTube, its Affiliates, officers, directors, employees and agents, from and against any and all claims, damages, obligations, losses, liabilities, costs or debt, and expenses (including but not limited to attorney's fees) arising from: (i) your use of and access to the Service; (ii) your violation of any term of this Agreement; (iii) your violation of any third party right, including without limitation any copyright, property, or privacy right; or (iv) any claim that your Content caused damage to a third party. This defense and indemnification obligation will survive this Agreement and your use of the Service.
YouTube — Version 73, retrieved January 31, 2026
Meta Platforms — Meta Terms of Service
You agree to indemnify and hold Facebook, its subsidiaries, affiliates, officers, agents, and other partners and employees, harmless from any loss, liability, claim, or demand, including reasonable attorney's fees, made by any third party due to or arising out of your use of the Service in violation of this Agreement or your violation of any law or the rights of a third party.
Meta (Facebook) — Version 1271, retrieved November 26, 2005
TikTok / ByteDance — TikTok Terms of Service
TikTok — Version 2265, retrieved March 30, 2021You warrant that any such contribution does comply with those standards, and you will be liable to us and indemnify us for any breach of that warranty. This means you will be responsible for any loss or damage we suffer as a result of your breach of warranty.
Any User Content will be considered non-confidential and non-proprietary. You must not post any User Content on or through the Services or transmit to us any User Content that you consider to be confidential or proprietary. When you submit User Content through the Services, you agree and represent that you own that User Content, or you have received all necessary permissions, clearances from, or are authorised by, the owner of any part of the content to submit it to the Services, to transmit it from the Services to other third party platforms, and/or adopt any third party content.
Alphabet / Google — Google Terms of Service
To the extent allowed by applicable law, you’ll indemnify Google and its directors, officers, employees, and contractors for any third-party legal proceedings (including actions by government authorities) arising out of or relating to your unlawful use of the services or violation of these terms or service-specific additional terms. This indemnity covers any liability or expense arising from claims, losses, damages, judgments, fines, litigation costs, and legal fees.
Alphabet (Google) — Version 3, retrieved January 31, 2026
Compare indemnification clauses across all Tech companies →
4. Arbitration and Class-Action Waivers
Mandatory arbitration clauses with class-action waivers are a critical barrier to the current litigation wave. While the MDL has proceeded in federal court, individual users who accepted these terms may face arguments that their claims must be arbitrated individually rather than pursued as a class.
Snap Inc. — Snapchat Terms of Service
ARBITRATION NOTICE: THESE TERMS CONTAIN AN ARBITRATION CLAUSE A LITTLE LATER ON. EXCEPT FOR CERTAIN TYPES OF DISPUTES MENTIONED IN THAT ARBITRATION CLAUSE, YOU AND SNAP INC. AGREE THAT DISPUTES BETWEEN US WILL BE RESOLVED BY MANDATORY BINDING ARBITRATION, AND YOU AND SNAP INC. WAIVE ANY RIGHT TO PARTICIPATE IN A CLASS-ACTION LAWSUIT OR CLASS-WIDE ARBITRATION.
Snap — Version 844727, retrieved October 13, 2017
TikTok / ByteDance — TikTok Terms of Service
ARBITRATION NOTICE FOR USERS IN THE UNITED STATES: THESE TERMS CONTAIN AN ARBITRATION CLAUSE AND A WAIVER OF RIGHTS TO BRING A CLASS ACTION AGAINST US. EXCEPT FOR CERTAIN TYPES OF DISPUTES MENTIONED IN THAT ARBITRATION CLAUSE, YOU AND TIKTOK AGREE THAT DISPUTES BETWEEN US WILL BE RESOLVED BY MANDATORY BINDING ARBITRATION, AND YOU AND TIKTOK WAIVE ANY RIGHT TO PARTICIPATE IN A CLASS-ACTION LAWSUIT OR CLASS-WIDE ARBITRATION.
TikTok — Version 2265, retrieved March 30, 2021
Meta Platforms — WhatsApp Terms of Service
Federal Arbitration Act. La Federal Arbitration Act des États-Unis régit l'interprétation et l'application du présent paragraphe « Disposition particulière concernant l'arbitrage pour les utilisateurs des États-Unis ou du Canada », y compris toute question déterminant si un Litige entre WhatsApp et vous est soumis à un arbitrage.
WhatsApp — Version 2303, retrieved December 31, 2022
Finding 3
Snap, TikTok, and WhatsApp all include mandatory binding arbitration provisions. Snap and WhatsApp explicitly include class-action waivers within their arbitration clauses — Snap states users "waive the right to participate in a class-action lawsuit or class-wide arbitration," and WhatsApp states users "WAIVE YOUR RIGHT TO PARTICIPATE IN CLASS ACTIONS, CLASS ARBITRATIONS, OR REPRESENTATIVE ACTIONS." TikTok's arbitration clause routes disputes to the Singapore International Arbitration Centre. These provisions, if enforceable, would force individual dispute resolution rather than collective litigation — dramatically increasing the cost of pursuing claims and reducing potential liability exposure for the platforms.
Compare arbitration clauses across all Tech companies → Compare class-action waiver clauses →
5. The Algorithm as "Service" — Not Third-Party Content
This is the structural move that makes the liability architecture work. Every platform defines what it provides to users as a single, undifferentiated "Service." The recommendation algorithm — the system that selects, ranks, and delivers content to each user — is part of that Service. It is not carved out as "third-party content." It is not characterized as "user speech." It is not treated as a separate product with its own warranty or liability terms.
This matters because Section 230 of the Communications Decency Act immunizes platforms from liability for third-party content. The current litigation argues that the algorithm is not third-party content — it is the platform's own product. And the Terms of Service implicitly agree: the algorithm is defined as part of the platform's proprietary "Service," subject to the platform's blanket liability limitations and damage caps.
The result is a contractual no-man's-land. When it comes to Section 230, platforms characterize themselves as passive intermediaries hosting user speech. But in their own Terms of Service, the recommendation engine is not "user speech" at all — it is the company's "Service," with liability capped and damages excluded. The platforms' contractual self-description — defining the algorithm as their own proprietary "Service" — complicates the purely passive-intermediary framing often advanced in Section 230 litigation.
A striking secondary pattern: while marketing and product documentation extensively describe these recommendation algorithms — Google details its "personalized search results, content, and ads," Meta describes how it "personalizes your experience" — the Terms of Service never mention the word "algorithm" in the context of liability. The algorithm is buried inside the larger "Service" definition, unnamed and unacknowledged.
Finding 4
Every platform contractually defines its recommendation algorithm as part of its "Service" — not as third-party content, not as user speech, and not as a separate product. No platform's Terms of Service contains a specific liability disclaimer for its recommendation algorithm, content-ranking system, or "For You" feed. Instead, all platforms rely on blanket liability limitations that cover the entire "Service." The same clause that caps liability for a server outage caps liability for an algorithm that surfaces harmful content to a thirteen-year-old. This creates a tension with Section 230 defenses: platforms define the algorithm as their own Service in their contracts, but may argue it is merely a conduit for third-party speech in court.
6. Damage Exclusions and the Harm Gap
The liability limitation clauses shown in Section 2 do double duty: they don't merely cap dollar amounts — they categorically exclude the types of damages most relevant to the algorithmic harm litigation. Emotional distress, mental health injuries, and loss of wellbeing are all consequential or indirect damages, which every platform's clause explicitly excludes.
Meta Platforms — Meta Terms of Service
Meta's limitation excludes liability for "any lost profits, revenues, information, or data, or consequential, special, indirect, exemplary, punitive, or incidental damages arising out of or related to these Terms or the Meta Products":
EXCEPT IN JURISDICTIONS WHERE SUCH PROVISIONS ARE RESTRICTED, IN NO EVENT WILL FACEBOOK BE LIABLE TO YOU OR ANY THIRD PERSON FOR ANY INDIRECT, CONSEQUENTIAL, EXEMPLARY, INCIDENTAL, SPECIAL OR PUNITIVE DAMAGES, INCLUDING ALSO LOST PROFITS ARISING FROM YOUR USE OF THE WEB SITE OR THE SERVICE, EVEN IF FACEBOOK HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. NOTWITHSTANDING ANYTHING TO THE CONTRARY CONTAINED HEREIN, FACEBOOK'S LIABILITY TO YOU FOR ANY CAUSE WHATSOEVER, AND REGARDLESS OF THE FORM OF THE ACTION, WILL AT ALL TIMES BE LIMITED TO THE AMOUNT PAID, IF ANY, BY YOU TO FACEBOOK FOR THE SERVICE DURING THE TERM OF MEMBERSHIP.
Meta (Facebook) — Version 1271, retrieved November 26, 2005
Snap Inc. — Snapchat Terms of Service
Snap excludes liability for "any indirect, incidental, special, consequential, punitive, or multiple damages":
Snap — Version 844727, retrieved October 13, 201716. Limitation of Liability
TO THE MAXIMUM EXTENT PERMITTED BY LAW, SNAP INC. AND OUR MANAGING MEMBERS, SHAREHOLDERS, EMPLOYEES, AFFILIATES, LICENSORS, AGENTS, AND SUPPLIERS WILL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, CONSEQUENTIAL, PUNITIVE, OR MULTIPLE DAMAGES, OR ANY LOSS OF PROFITS OR REVENUES, WHETHER INCURRED DIRECTLY OR INDIRECTLY, OR ANY LOSS OF DATA, USE, GOODWILL, OR OTHER INTANGIBLE LOSSES, RESULTING FROM: (A) YOUR ACCESS TO OR USE OF OR INABILITY TO ACCESS OR USE THE SERVICES; (B) THE CONDUCT OR CONTENT OF OTHER USERS OR THIRD PARTIES ON OR THROUGH THE SERVICES; OR (C) UNAUTHORIZED ACCESS, USE, OR ALTERATION OF YOUR CONTENT, EVEN IF SNAP INC. HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. IN NO EVENT WILL SNAP INC.’S AGGREGATE LIABILITY FOR ALL CLAIMS RELAT...
Finding 5
Every platform categorically excludes liability for consequential, indirect, incidental, and punitive damages. The mental health harms alleged in the algorithmic addiction litigation — anxiety, depression, eating disorders, self-harm — would likely be classified as consequential or indirect damages under these clauses. The contractual framework attempts to foreclose exactly the category of injury that plaintiffs allege, regardless of whether those injuries were caused by third-party content or by the platform's own recommendation algorithm.
7. Temporal Evolution: How These Clauses Have Changed
Where our dataset includes multiple historical versions of a platform's Terms of Service, we can observe how liability-related language has evolved. These observations are limited to the versions in our dataset and may not capture every change.
Meta / Facebook: 2005 → 2026
We have Facebook's original 2005 Terms of Service alongside Meta's current terms, spanning the full arc of the platform's evolution from college directory to algorithmic feed.
Scope expansion: The 2005 limitation of liability covered "YOUR USE OF THE WEB SITE OR THE SERVICE." The current version covers "these Terms or the Meta Products." The shift from singular "Service" and "Web site" to plural "Meta Products" reflects the expansion of what the terms cover — a category that includes the News Feed algorithm, Reels recommendation engine, and all personalization systems, none of which existed in 2005.
Indemnification narrowing: The 2005 indemnification clause required users to hold Facebook harmless for "your use of the Service in violation of this Agreement or your violation of any law or the rights of a third party." The current Meta Terms of Service do not contain a separately extracted indemnification clause — liability allocation is handled entirely through the limitation-of-liability and forum-selection provisions.
View Facebook 2005 Terms → View Meta 2026 Terms →
Snap: "Even If Recommended" — Added August 27-28, 2021
Using the Wayback Machine, we pinpointed exactly when Snap added the critical "even if recommended" language to its indemnification clause. The change went live between August 27, 2021 (5:49 PM UTC) and August 28, 2021 (1:51 AM UTC).
For context, several events occurred in the months before the change. On May 4, 2021, the Ninth Circuit ruled in Lemmon v. Snap, Inc. that Snap does not enjoy Section 230 immunity for product-design claims — a landmark ruling that exposed the company to design-defect liability for the first time. On July 21, 2021, eleven-year-old Selena Rodriguez of Enfield, Connecticut died by suicide after what a treating therapist described as the most severe social media addiction she had ever seen; a subsequent wrongful death lawsuit filed in January 2022 named both Snap and Meta. The indemnification change occurred 115 days after the Lemmon ruling and 37 days after Rodriguez's death. We do not assert any causal connection between these events and the contract change.
The indemnification clause was rewritten from a standard three-prong structure to a four-prong structure with significantly expanded scope. The critical addition: users indemnify Snap for products or services "even if recommended, made available, or approved by Snap."
Timeline
May 4, 2021
Lemmon v. Snap — 9th Circuit rules Snap has no Section 230 immunity for product design
July 21, 2021
Selena Rodriguez, 11, dies by suicide (Snap and Instagram addiction alleged)
Aug 27-28, 2021
Snap adds "even if recommended, made available, or approved by Snap" to indemnification clause
Oct 5, 2021
Frances Haugen testifies before Congress on Facebook/Instagram harm to children
Jan 20, 2022
Rodriguez wrongful death lawsuit filed against Snap and Meta
Before — August 25, 2021
...relating in any way to: (a) your access to or use of the Services; (b) your content; and (c) your breach of these Terms.
After — August 28, 2021
...relating in any way to: (a) your access to or use of the Services, or any products or services provided by a third party in connection with the Services, even if recommended, made available, or approved by Snap; (b) your content, including infringement claims related to your content; (c) your breach of these Terms or any applicable law or regulation; or (d) your negligence or willful misconduct.
Finding 6: Significant Expansion
Snap's August 2021 indemnification rewrite was not a minor update. Four changes were made simultaneously: (1) users indemnify Snap for third-party products recommended by Snap, (2) content indemnification was expanded to include infringement claims, (3) the breach trigger was broadened from "these Terms" to "any applicable law or regulation," and (4) a new prong was added covering "your negligence or willful misconduct." The "even if recommended" language is notable because it covers the scenario at issue in the algorithmic addiction litigation — harm flowing from content that Snap's algorithm recommended — and contractually assigns that risk to the user.
We do not claim to know Snap's internal motivations for this change, and we do not assert that it was made in response to any particular event. We note the factual timeline: the change occurred 115 days after the Ninth Circuit ruled in Lemmon v. Snap that Snap lacks Section 230 immunity for product-design claims, and 37 days after the death of Selena Rodriguez, an 11-year-old whose therapist described her as the most severely social-media-addicted patient she had ever treated. The scope and timing of the change are matters of public record; readers may draw their own conclusions.
TikTok: Regional Variants
TikTok maintains distinct Terms of Service for different regions. The US version (TikTok USDS) explicitly provides the platform "as is" and mentions "generative AI-enabled features" in its liability clause. The international version uses a UK-style limitation focused on excluding liability for death and personal injury caused by negligence. The US-specific "as is" language and the explicit mention of AI features are notable — they suggest awareness of the product-liability framing.
WhatsApp: 2020 → 2021 → 2026
WhatsApp has the deepest version history in our dataset (8 versions from 2020-2026). The December 2020 version contained a detailed limitation of liability clause explicitly excluding "LOST PROFITS OR CONSEQUENTIAL, SPECIAL, PUNITIVE, INDIRECT, OR INCIDENTAL DAMAGES." The arbitration clause gained the prominent all-caps class-action waiver header visible in the current version. These changes coincide with the period when the algorithmic addiction litigation was gaining momentum.
Temporal Limitations
Our historical coverage varies by platform: Meta/Facebook spans 2005-2026, Snap covers 2021-2026 with the indemnification change pinpointed to a specific date, and WhatsApp has 8 versions from 2020-2026. For other platforms, our dataset contains 1-2 versions, limiting temporal analysis.
8. Implications for Litigation
The contractual framework across these platforms creates a multi-layered liability shield:
- "Service" definition: The algorithm is contractually part of the platform's "Service" — not third-party content, not user speech — buried in a unitary product definition with no separate terms.
- Damage exclusions: Consequential, indirect, and punitive damages — the categories covering mental health harm — are categorically excluded.
- Liability cap: Even if liability survives the exclusions, damages are typically capped at $100 or the amount paid in the prior 12 months (effectively $0 for free services).
- Indemnification: Users agree to cover the platform's legal costs arising from the user's use of the service.
- Arbitration + class waiver: Claims must be pursued individually in arbitration, not in court as a class action.
Whether these provisions are enforceable against minors — who are the primary plaintiffs in the current litigation — is a central question. Minors generally have the right to disaffirm contracts, and courts have historically been skeptical of enforcing arbitration agreements against children. Several state attorneys general have also argued that these provisions are unconscionable as applied to minors.
Litigation Uses
This contractual record supports at least three litigation strategies:
- Design defect / product architecture framing (Section 230 avoidance): Because the platforms themselves define the recommendation algorithm as part of their proprietary "Service" — not as third-party content — plaintiffs can argue the algorithm is the company's own product, not protected speech. The Terms of Service are the company's own admission that the algorithm is its product, delivered on its terms, under its warranties (or lack thereof). This undermines the Section 230 defense that the platform is merely hosting third-party content.
- Unconscionability / minors / public policy: These contracts are take-it-or-leave-it adhesion agreements offered to children as young as 13. They exclude liability for consequential and indirect damages (the categories covering mental health harm), cap remaining damages at $0, require binding arbitration, and waive class actions — for a product the company designed and controls entirely. Courts applying unconscionability doctrine, state consumer protection statutes, or public policy limits on contracting with minors may find this combination unenforceable.
- Foreseeability and intentional risk allocation (Snap's recommendation-indemnity clause): Snap's indemnification clause covers products and services "even if recommended, made available, or approved by Snap". This is not boilerplate — it specifically anticipates the scenario where Snap's recommendation causes harm and contractually shifts that risk to the user. This language is evidence of foreseeability: the company anticipated algorithmic recommendation claims and drafted contract language to allocate that risk away from itself, which cuts against any defense that harm from recommendation algorithms was unforeseeable.
Appendix: Auditing the "Algorithm Not Mentioned" Claim
Section 5 asserts that no platform mentions the word "algorithm" in the context of liability. This claim is auditable. The links below search the full text of each platform's current Terms of Service for the words "algorithm," "recommendation," and "personalize." In every case, these words either do not appear at all in the Terms of Service, or appear only in descriptive or marketing contexts — never in a warranty disclaimer, liability cap, indemnification clause, or damage limitation.
| Platform | Search "algorithm" | Search "recommend" | Search "personalize" |
|---|---|---|---|
| Meta — Terms of Service | Search → | Search → | Search → |
| Facebook — Terms of Service | Search → | Search → | Search → |
| WhatsApp — Terms of Service | Search → | Search → | Search → |
| Google — Terms of Service | Search → | Search → | Search → |
| YouTube — Terms of Service | Search → | Search → | Search → |
| Snap — Terms of Service | Search → | Search → | Search → |
| TikTok — Terms of Service | Search → | Search → | Search → |
Each link opens the full text of that version. Use your browser's Find (Ctrl+F / Cmd+F) to verify the search term's presence or absence. Where these terms do appear, note whether they are in a liability-related clause or in descriptive/product language.
Methodology & Limitations
- Scope: This case study examines Terms of Service only. Privacy Policies, Community Guidelines, and supplemental terms may contain additional relevant language.
- Clause extraction: Clauses were extracted using automated pattern matching. Some provisions may be incomplete or missing. All extracted text links to the source version for verification.
- Temporal depth: Our historical coverage varies by platform. Some documents have only 1-2 tracked versions. We cannot make claims about when specific language was added without deeper historical data.
- Not legal analysis: This is a descriptive study of contractual language, not a legal opinion on enforceability. Contract enforceability depends on jurisdiction, the age and capacity of the user, unconscionability doctrine, and applicable statutory protections.
- Data sources: All text is drawn from publicly available Terms of Service, retrieved and archived by TOS Tracker. See our Methodology page for details on how we collect and process documents.
- Live data: This page queries TOS Tracker’s database on every load. As new documents are added to our corpus or existing documents are updated, the clause excerpts, platform comparisons, and findings on this page update automatically. No manual curation is required.
Reproduce this analysis:
Revision History
If citing this case study, please include the revision date to ensure readers can locate the version you referenced.
| Date | Change |
|---|---|
| February 11, 2026 | Initial publication. |
| February 11, 2026 | Revised Section 230 framing in Section 5 to avoid overclaiming; softened temporal language throughout; added author byline and PDF export. |
Suggested citation: Leahey, Andrew. "Algorithmic Liability: How Social Media Platforms Disclaim Responsibility for Their Recommendation Algorithms." TOS Tracker, February 11, 2026. https://tostracker.app/analysis/algorithmic-liability