Provisions index

Before any AI-generated recommendation in a designated high-risk domain (legal, medical, financial, credit, insurance, housing, employment, education, healthcare access, public benefits, legal status) reaches an affected individual, a qualified human with relevant expertise and applicable licensure must review and have authority to modify that recommendation. The reviewing person retains accountability for the final recommendation....

Why it matters: This provision establishes an affirmative operational obligation on operators and developers deploying Claude in twelve enumerated high-stakes categories, requiring documented human oversight workflows before AI output is acted upon or delivered. The clause specifies licensure requirements by reference to applicable law, meaning compliance exposure varies by jurisdiction and professional domain....

View provision → Watch Anthropic →

Operators must disclose to affected individuals that AI was involved in producing any High-risk AI Recommendation delivered to them or used to make a decision about them, though they are not required to identify the specific AI system or provider....

Why it matters: This provision establishes a mandatory disclosure obligation at the point of AI-assisted decision delivery in high-risk domains, creating an operational requirement for operators to implement consumer-facing disclosure mechanisms in workflows involving legal, medical, financial, employment, housing, and related decisions....

View provision → Watch Anthropic →

The AUP prohibits using any outputs generated by Anthropic's models to train other AI models, including through distillation or scraping techniques, unless the user has received prior authorization from Anthropic....

Why it matters: This provision directly restricts a common developer workflow in which API outputs are used as training data for fine-tuning or building derivative models, and requires affirmative advance authorization from Anthropic before any such use is permitted....

View provision → Watch Anthropic →

Anthropic states that it will report detected CSAM, related facilitation activity, or imminent risk scenarios involving minors to relevant authorities, and applies a uniform definition of minor as any person under 18 regardless of jurisdiction....

Why it matters: This provision discloses Anthropic's affirmative practice of reporting CSAM-related detections to law enforcement or equivalent authorities, and establishes a platform-wide age definition of 18 regardless of local age of majority definitions, which has compliance implications for operators deploying Claude in jurisdictions with differing age-related standards....

View provision → Watch Anthropic →

The AUP prohibits using Claude to make or recommend decisions in criminal justice processes, to conduct non-consensual surveillance of individuals including via facial recognition or predictive policing, and to build or enhance surveillance infrastructure for these prohibited uses....

Why it matters: This provision categorically excludes a range of law enforcement, criminal justice, and surveillance applications from permitted use, establishing explicit prohibitions on facial recognition, battlefield management, predictive policing, and non-consensual individual tracking as distinct named examples....

View provision → Watch Anthropic →

When Claude's outputs control hardware capable of physical harm (including autonomous vehicles, surgical devices, industrial machinery, and hazardous materials systems), a qualified human observer must maintain real-time ability to halt operation, and all physical safety limits must be enforced by hardware or software independent of the AI model's output....

Why it matters: This provision establishes hardware-level architectural requirements for AI-physical system integrations, mandating that safety limits be enforced outside the model control loop and that fail-safe behavior occur upon loss of connection to Anthropic's services, which has direct implications for robotics, medical device, industrial automation, and autonomous vehicle deployments....

View provision → Watch Anthropic →

Operators deploying consumer-facing chatbots or AI agents built on Claude must disclose to users at the start of each chat session, or in the product interface, that they are communicating with AI rather than a human; the disclosure need not name Anthropic or the specific model....

Why it matters: This provision establishes a session-level or interface-level disclosure obligation for all consumer-facing Claude-powered chatbot deployments, creating an operational requirement for operators to implement disclosure mechanisms in product UI or chat initiation flows....

View provision → Watch Anthropic →

Anthropic reserves the right to warn, throttle, limit, suspend, or terminate any user's access upon suspected AUP violation, and operates real-time output safeguards that may block specific outputs independently of whether a formal AUP violation has occurred....

Why it matters: This provision establishes that enforcement action may be taken on suspicion of violation without a confirmed finding, and that real-time blocks do not constitute AUP violations, distinguishing between automated safeguard responses and formal enforcement decisions....

View provision → Watch Anthropic →

The AUP categorically prohibits using Claude at any stage of CBRN weapons development, including synthesis, design, acquisition, testing, or deployment planning for biological, chemical, radiological, or nuclear weapons, their precursors, or delivery systems, as well as modification of biological or chemical agents to enhance harm potential....

Why it matters: This provision establishes an absolute prohibition covering the full lifecycle of CBRN weapons development activity and explicitly names weapons-critical technology categories such as fissile material production and enrichment, creating a categorical exclusion for defense and dual-use research operators in these domains....

View provision → Watch Anthropic →

Operators who build or deploy agentic AI systems using Claude are responsible for ensuring that all actions those agents take through connected tools, browsers, or external systems comply with the AUP, extending operator accountability to agent-initiated downstream actions....

Why it matters: This provision allocates AUP compliance responsibility to the operator for the full scope of actions taken by Claude-powered agents in agentic deployments, including tool use, browser interactions, and connected system actions, which extends compliance obligations beyond direct model prompting to autonomous agent behavior....

View provision → Watch Anthropic →

The policy states that Claude conversation inputs and outputs may be used to train Anthropic's AI models, with an opt-out available in account settings. The opt-out does not apply to content that has been flagged for safety review or that the user has explicitly reported through feedback mechanisms....

Why it matters: This provision establishes a conditional opt-out structure in which the stated right to exclude conversational data from model training is subject to exceptions that activate when content is flagged by Anthropic's systems or reported by the user. Compliance teams should assess whether this structure satisfies informed consent and data minimization requirements under GDPR and CCPA, particularly whether the exceptions are disclosed with sufficient prominence at the point of opt-out....

View provision → Watch Anthropic →

The policy discloses that Anthropic may collect facial geometry templates as part of identity or age verification processes, and acknowledges these may constitute biometric data in some jurisdictions. Collection is described as occurring only when users choose to verify their identity....

Why it matters: This provision establishes a basis for collecting biometric identifiers including facial geometry templates, which are subject to heightened regulatory requirements in multiple U.S. states and internationally. The policy's acknowledgment that this data 'may be considered biometric data in some jurisdictions' indicates awareness of jurisdictional variation but does not specify which consent or notice obligations are triggered in each relevant jurisdiction....

View provision → Watch Anthropic →

The policy states that using a work or organizational email address to create an Anthropic account may result in the account being linked to that organization's enterprise account, potentially making account data accessible to organizational administrators....

Why it matters: This provision establishes that individual consumer accounts created with employer-owned email addresses may be administratively linked to enterprise accounts, which could affect the privacy expectations of individual employees and require organizations to assess their data governance obligations with respect to employee use of Claude.ai....

View provision → Watch Anthropic →

The policy states that when third-party integrations are enabled, Claude may transmit inputs, outputs, and instructions to those third-party services, which then process the data under their own privacy policies. Anthropic expressly disclaim responsibility for third-party data practices. Some integrations may enable ongoing third-party access until the user disables them....

Why it matters: This provision establishes that enabling third-party integrations results in data flows to third parties outside Anthropic's privacy policy coverage, with ongoing access possible until the user affirmatively disables the integration. The policy places responsibility for reviewing third-party privacy policies on users prior to enabling integrations....

View provision → Watch Anthropic →

The policy states that Anthropic's services are not directed at users under 18 and that Anthropic does not knowingly collect data from minors under 18. The policy states that detection and removal measures are in place and provides a reporting mechanism for known minor data collection....

Why it matters: This provision establishes a minimum age of 18 for use of Anthropic's consumer services, which is above the 13-year threshold established by COPPA for U.S. platforms. The policy's statement that detection measures are in place may be relevant to regulatory assessments of whether Anthropic's age assurance practices are adequate under applicable law....

View provision → Watch Anthropic →

The policy establishes a rights exercise process requiring identity verification and, for third-party representatives, evidence of authorization. Response timelines are stated to follow applicable law, with GDPR and UK GDPR requests acknowledged within one calendar month and extendable by two months for complex requests....

Why it matters: This provision establishes the procedural requirements for exercising data subject rights under Anthropic's policy, including identity verification requirements that apply to all requests and authorization evidence requirements for third-party representatives. The policy notes that processes for acting on training dataset requests are complex, which may affect the practical ability to fulfill deletion or correction requests involving data used in AI model training....

View provision → Watch Anthropic →

The guidelines require all users who post content that promotes a product, brand, or business to use TikTok's proprietary content disclosure setting; this obligation applies regardless of whether the promotion is paid or organic....

Why it matters: This provision requires affirmative use of a platform-specific disclosure mechanism for any promotional content, establishing a compliance obligation that interacts with FTC endorsement and disclosure guidance applicable to influencers, creators, and business accounts operating in the US market....

View provision → Watch TikTok →

The guidelines require creators to apply clear labels to content where AI or editing tools are used to realistically depict people or scenes, and prohibit AI-generated content that misleads on matters of public importance or harms individuals....

Why it matters: This provision establishes an affirmative disclosure obligation for AI-assisted content depicting people or scenes, and creates a separate prohibition on misleading AI-generated content, with both requirements subject to platform enforcement including content removal....

View provision → Watch TikTok →

The guidelines establish three operationally distinct enforcement responses to content: outright removal for prohibited content, age-restriction limiting viewership to adults for certain content, and exclusion from the For You Feed recommendation system without removal for other content categories....

Why it matters: The three-tier system means that content may be subject to reduced distribution through FYF exclusion without triggering account-level enforcement or removal, creating a distinct operational outcome that affects creator reach and monetization without constituting a formal violation finding....

View provision → Watch TikTok →

The guidelines prohibit trade, marketing, or facilitation of access to regulated, prohibited, or high-risk goods and services for general users, but create an exception permitting verified business accounts and verified TikTok Shop sellers to sell or market certain regulated items subject to criteria described only as 'strict requirements.'...

Why it matters: The carve-out for verified sellers creates a tiered access system for regulated goods categories, but the specific eligibility criteria are not disclosed in this document, creating a disclosure gap for sellers seeking to understand their compliance obligations and eligibility status....

View provision → Watch TikTok →

The guidelines prohibit misinformation that could cause significant harm to individuals or society, and specifically prohibit content that could mislead voters or interfere with elections, including false claims about voting procedures, eligibility, or election outcomes....

Why it matters: These provisions establish platform-level content removal authority over a broad category of information-related content, with the operative threshold of 'significant harm' undefined in the document, creating interpretive discretion in enforcement that may affect political, journalistic, and public interest content....

View provision → Watch TikTok →

The guidelines prohibit sharing personal information that could lead to harms including identity theft, stalking, or fraud, and separately prohibit any attempts to hack, reverse-engineer, or compromise TikTok's platform systems....

Why it matters: These provisions establish content removal authority over doxing and related privacy-harming disclosures, and create a platform-level prohibition on security research or reverse-engineering activities, the latter of which may interact with security researcher safe harbor expectations and applicable computer fraud statutes....

View provision → Watch TikTok →

The guidelines prohibit accounts that engage in deceptive behavior or attempt to manipulate the platform, and prohibit the trade of services designed to artificially boost engagement metrics or circumvent the recommendation algorithm....

Why it matters: This provision creates account-level enforcement exposure for users who purchase or sell engagement-boosting services, and applies to both the accounts purchasing such services and those providing them, with implications for influencer marketing practices and platform eligibility....

View provision → Watch TikTok →

The document explicitly states that the erotic and violent speech content moderation mitigations do not cover nonverbal vocalizations or sound effects such as erotic moans, violent screams, or gunshots; some general refusal behavior for sound effects exists but is not the product of a designed mitigation....

Why it matters: This provision constitutes an explicit documented disclosure of a known gap in content moderation coverage at the time of deployment. The absence of a designed mitigation for nonverbal harmful content creates a residual risk that is acknowledged in the system card, which compliance teams and API integrators should factor into their own risk assessments and downstream content policies....

View provision → Watch OpenAI →

The document discloses that at time of deployment, mitigations for audio robustness degradation under real-world conditions (background noise, interruptions, low audio quality) and for audio-delivered misinformation and conspiracy theories are nascent or still in development; red teamers successfully elicited misinformation via audio prompts....

Why it matters: This provision constitutes a disclosed residual risk at deployment, establishing that the model was released with known, documented safety limitations in audio modalities for which mitigations were not yet complete. The combination of audio delivery and emotive or emphatic speech style is identified as a factor that may increase the persuasiveness of misinformation, per red teamer observations....

View provision → Watch OpenAI →

The provision states that using Kraken's services or accepting its privacy notice constitutes written consent to biometric data collection and processing, and operates as a release of both past and future claims related to that biometric data handling....

Why it matters: This provision asserts a prospective and retrospective claims release tied to biometric data processing, which may interact with state biometric privacy statutes such as the Illinois Biometric Information Privacy Act (BIPA), which provides a private right of action and under which courts have examined whether contractual claim waivers are enforceable. Compliance teams operating in BIPA-covered jurisdictions should evaluate whether this release mechanism aligns with applicable state law requirements....

View provision → Watch Kraken →

The GLBA Notice discloses that Kraken shares personal information with affiliates for everyday business purposes and marketing, and authorizes marketing by affiliates, with opt-out rights available for affiliate marketing sharing and creditworthiness sharing but not for sharing for everyday business purposes or affiliates' everyday business purposes regarding transactions and experiences....

Why it matters: This provision establishes the scope of permissible financial information sharing under GLBA and specifies which sharing categories are subject to consumer opt-out. Sharing for everyday business purposes, including with affiliates for transaction and experience information, is stated as non-restrictable, while affiliate marketing sharing and creditworthiness sharing with affiliates are opt-outable....

View provision → Watch Kraken →

The provision states that Kraken shares users' Social Security numbers, date of birth, name, phone number, and ZIP code with third-party payroll-connectivity partners when users access payroll-linked services such as direct deposit switching....

Why it matters: This provision authorizes sharing of a combination of highly sensitive identifiers, including Social Security numbers, with named third-party payroll-connectivity partners for the purpose of locating payroll records. This data combination represents a category with significant identity-related sensitivity and should be subject to vendor due diligence and contractual data use limitations....

View provision → Watch Kraken →

The provision states that residents of nineteen named states have privacy rights including access, deletion, correction, portability, and opt-out of targeted advertising, sales, and profiling, with specific exclusions for Iowa and Utah regarding correction and profiling opt-out rights....

Why it matters: This provision acknowledges a multi-state privacy rights framework applicable to residents of nineteen states, establishing that Kraken recognizes and will process rights requests under these laws. The provision's acknowledgment that rights may vary by state, and its specific exclusions for Iowa and Utah, indicates operational differentiation in rights response workflows....

View provision → Watch Kraken →

The policy authorizes Stash to use personal information collected from users to train large language models, disclosed in a single sentence in Section 4 without specifying which data categories are used, retention periods for training purposes, or available opt-out mechanisms....

Why it matters: This provision authorizes a data use case beyond core service delivery without elaborating on scope, applicable data categories, or opt-out availability, which may require evaluation under CCPA business purpose disclosure requirements and emerging state AI and automated decision-making regulations....

View provision → Watch Stash →

The terms grant Stash and Plaid authority to access and transmit personal and financial information, including historical and ongoing transaction data, from all sub-accounts accessible through any linked bank credential set, including sub-accounts the user did not specifically designate for linking....

Why it matters: This provision establishes that account linking authorizes data access beyond the single sub-account designated by the user, extending to all sub-accounts accessible through the same credential set, including checking, savings, and credit card accounts; this scope of financial data access may require evaluation under GLBA data minimization principles and applicable state financial privacy statutes....

View provision → Watch Stash →

The policy authorizes Stash to draw inferences from collected personal information to create profiles that may reflect psychological trends, predispositions, attitudes, intelligence, abilities, and aptitudes, in addition to behavioral and preference characteristics....

Why it matters: This provision authorizes inference-drawing across the full scope of personal information collected by Stash, including sensitive financial identifiers and transaction data, to construct profiles that extend to psychological and attitudinal attributes; this scope of profiling may require evaluation under state automated decision-making statutes and CCPA sensitive personal information provisions....

View provision → Watch Stash →

The policy states that state-law privacy rights such as data access, deletion, and correction do not apply to Nonpublic Personal Information that is subject to GLBA, which includes financial account data, transaction records, and other information collected in connection with financial products or services....

Why it matters: This provision establishes that a significant portion of the personal information Stash collects from financial services customers, specifically Nonpublic Personal Information as defined in Section 2.2, falls outside the scope of state privacy rights frameworks, which may limit the practical scope of consumer data access and deletion requests for financial customers....

View provision → Watch Stash →

The policy states that by providing a phone number, users consent to automated and prerecorded calls and texts from Stash, and acknowledges that users may incur charges for receiving such communications; consent to informational text messages may be revoked by replying STOP....

Why it matters: This provision conditions platform phone number provision on consent to automated dialing and prerecorded messaging under the TCPA, and the terms acknowledge the possibility that users may incur costs from receiving these communications; the opt-out mechanism is limited to text messages and does not explicitly address revocation of consent to automated calls....

View provision → Watch Stash →

The policy reserves Stash's right to modify the privacy policy at any time, with material changes communicated by email or platform notice; continued use of the platform after changes constitutes acceptance, and if a user does not agree to changes, Stash reserves the right to terminate the relationship....

Why it matters: This provision establishes that continued use of the Stash platform following a policy change constitutes acceptance of revised terms, and separately states that Stash may terminate the user relationship if the user declines to accept future changes; for a financial services platform where account closure may affect ongoing investment positions, this termination right has operational significance....

View provision → Watch Stash →

The policy discloses that Stash collects sensitive personal information categories including government identification numbers, financial account credentials with access codes, precise geolocation, and citizenship or immigration status....

Why it matters: This provision identifies specific sensitive personal information categories subject to heightened regulatory treatment under CCPA and equivalent state statutes, including citizenship and immigration status, which may present distinct legal considerations in the context of government data requests or law enforcement disclosure....

View provision → Watch Stash →

Guideline 3.1.1 requires developers to use Apple's in-app purchase system as the exclusive payment mechanism for digital content, features, subscriptions, and in-game currencies, prohibiting alternative unlocking or payment mechanisms within the app. Regional exceptions exist for the US storefront and for developers holding specific entitlements in other designated storefronts....

Why it matters: This provision governs the monetization architecture of all digital goods sold through App Store apps, requiring payment to be processed through Apple's system and subjecting transactions to the Apple Developer Program commission structure. The regional carve-outs for the US storefront and entitlement-based external link permissions in designated regions reflect regulatory developments and create differentiated compliance obligations by geography....

View provision → Watch Apple →

Guideline 5.4 prohibits VPN apps from selling, using, or disclosing user data to third parties for any purpose, requires this commitment in the app's privacy policy, and mandates use of the NEVPNManager API. Non-compliance results in app removal from the App Store and blocking from alternative distribution....

Why it matters: This provision asserts an absolute prohibition on third-party data sharing for VPN apps, which extends beyond standard data minimization or purpose limitation requirements to bar any third-party data use regardless of purpose, including analytics, fraud prevention, or service provider relationships. The enforcement consequence includes blocking from alternative distribution channels, not only App Store removal....

View provision → Watch Apple →

Guideline 1.3 prohibits apps in the Kids Category from including third-party analytics or third-party advertising, with narrow exceptions for analytics that do not collect IDFA or identifiable child information and for contextual advertising services that publicly document human review of ad creatives for age appropriateness. Apps in this category must also comply with applicable children's privacy laws worldwide....

Why it matters: This provision establishes data collection and advertising restrictions for Kids Category apps that go beyond baseline COPPA compliance requirements by prohibiting third-party analytics and advertising as a default, limiting the monetization and measurement tools available to developers in this category. The exceptions require affirmative verification of third-party service compliance with specified criteria....

View provision → Watch Apple →

Guideline 3.2.2(ix) prohibits personal loan apps from charging an APR exceeding 36% inclusive of all costs and fees, and prohibits repayment terms of 60 days or less, while requiring conspicuous disclosure of all loan terms including APR and payment due date. This provision applies to apps distributed through the App Store....

Why it matters: This provision establishes a substantive APR cap and minimum loan term requirement that Apple applies as a condition of App Store distribution, which may affect fintech and consumer lending apps regardless of whether applicable state or federal law imposes equivalent restrictions in the jurisdictions where those apps operate....

View provision → Watch Apple →

Guideline 5.1.1(ii) requires apps to obtain user consent before collecting any user or usage data, including data considered anonymous at collection, and prohibits conditioning paid functionality on consent to data collection. Apps must provide a mechanism for users to withdraw consent and must describe data use clearly in purpose strings....

Why it matters: This provision requires consent for data collection regardless of whether the data is classified as anonymous at the time of collection, which is a broader consent trigger than some regulatory frameworks apply. The prohibition on conditioning paid functionality on data access consent establishes a platform-level data minimization standard applicable to all App Store apps....

View provision → Watch Apple →

Guideline 4.8 requires that any app using a third-party social login for primary account authentication must also offer an alternative login option that limits data collection to name and email, permits email privacy, and does not collect app interaction data for advertising without consent. Sign In with Apple satisfies these requirements....

Why it matters: This provision effectively requires that Sign In with Apple or an equivalent privacy-preserving login option be offered alongside any third-party social login used for primary account authentication, limiting the extent to which social platforms can capture user authentication data through App Store apps. Developers using multiple social login options must verify that at least one compliant alternative is available....

View provision → Watch Apple →

Guideline 5.1.3(i) prohibits apps from using or sharing health, fitness, and medical research data collected from specified Apple health APIs for advertising, marketing, or use-based data mining, and permits health research use only with explicit user permission. The prohibition extends to third-party disclosure for these purposes....

Why it matters: This provision establishes categorical restrictions on health data use that apply to all apps accessing Apple health APIs, independent of whether the app is a HIPAA-covered entity. The restriction on use-based data mining for health data is broader in its stated scope than standard commercial health data governance terms typically applied outside HIPAA-regulated contexts....

View provision → Watch Apple →

Guideline 5.1.1(viii) prohibits apps from aggregating or compiling personal information from third-party sources, including publicly available databases, without explicit user consent. This prohibition applies to both App Store distribution and alternative distribution channels....

Why it matters: This provision prohibits a category of data aggregation and people-search functionality that is common in data broker and background check applications, and extends the prohibition to alternative distribution channels beyond the App Store. The explicit consent requirement for compiling data from public databases is a substantive restriction on data aggregation business models....

View provision → Watch Apple →

The policy authorizes Google to associate activity collected across its own services and across third-party sites and apps that use Google products with a user's personal information, depending on account settings, for purposes including personalized advertising and service improvement....

Why it matters: This provision establishes the operational basis for Google's cross-service and cross-device advertising and personalization infrastructure, affecting how data collected across Search, YouTube, Gmail, and third-party integrated sites is linked to individual accounts....

View provision → Watch Google →

The policy states that Google uses publicly available information to train its AI models for products including Google Translate, Gemini Apps, and Cloud AI capabilities....

Why it matters: This provision discloses that publicly available information is used as training data for Google's AI systems, including Gemini Apps and Cloud AI, without specifying the categories of publicly available information collected or the consent mechanisms applicable to individuals whose publicly available data may be included....

View provision → Watch Google →

The policy discloses that third-party websites and apps using Google advertising or analytics services share user activity data with Google regardless of browser mode, including when users are browsing in Chrome Incognito mode....

Why it matters: This provision establishes that Incognito mode does not prevent third-party sites using Google services from sharing activity data with Google, which is operationally significant for users who rely on Incognito mode as a privacy measure when visiting third-party sites....

View provision → Watch Google →

The policy discloses that for users on organizational Google accounts, domain administrators and resellers have access to account data including email content, can change passwords, suspend account access, and can restrict the user's ability to delete or edit information or privacy settings....

Why it matters: This provision establishes that users operating under organizational Google accounts, including students using Google Workspace for Education and employees using Google Workspace, have reduced personal control over their account data and privacy settings relative to individual consumer accounts....

View provision → Watch Google →

The policy states that deleted data may persist in Google's active and backup systems for an unspecified period before complete removal, and that retained data may be kept in anonymized form rather than deleted entirely in some circumstances....

Why it matters: This provision establishes that user-initiated deletion requests do not result in immediate removal from all Google systems, and that some data may be retained in anonymized form, which has implications for compliance with data subject deletion rights under GDPR and U.S. state privacy laws....

View provision → Watch Google →

The policy authorizes collection of precise location data from GPS, IP address, Wi-Fi access points, cell towers, and Bluetooth-enabled devices, and states this data is used for service delivery and for advertising based on a user's location....

Why it matters: This provision establishes that Google collects multiple categories of location data, potentially including precise GPS coordinates as well as inferred location from network infrastructure, and authorizes use of that data for targeted advertising, subject to device and account settings....

View provision → Watch Google →