Grok vs. Meta AI: data protection per plan

no (DPA is part of the Enterprise terms for business customers; consumer use is subject to the Consumer terms)

no (DPA is part of the Enterprise terms for business customers; consumer use is subject to the Consumer terms)

yes: Enterprise FAQ: 'if you submit personal data to our Services, our Data Processing Addendum (DPA) applies and is automatically incorporated into our Enterprise Terms of Service'; the Grok Business guide refers, for team workspaces, to the 'Terms of Service - Enterprise', which according to their preamble also apply to purchases by 'online purchase confirmation'. The DPA (effective 2026-09-22) provides for the processing of personal data only via the ZDR API.

yes: Data Processing Addendum (effective 2026-09-22) is part of the Enterprise terms for business services; the DPA provides for the processing of personal data only via the ZDR API.

yes: Data Processing Addendum (effective 2026-09-22) is part of the 'Terms of Service - Enterprise', which according to a notice apply to business customers of the API and other business services; SCCs Module Two/Three. The DPA requires personal data to be processed only via the ZDR API; no personal data is to be transmitted to non-ZDR services.

yes: Data Processing Addendum (effective 2026-09-22) as part of the Enterprise terms; the DPA expressly provides for the processing of personal data only via the ZDR API. SCCs Module Two/Three (Irish law).

not publicly documented (the Meta AI Terms of Service (EU) are a consumer contract with Meta Platforms Ireland Limited and refer to Meta's privacy policy for personal data; processor terms are not mentioned there)

yes, if Contributor mode is switched off: section 5 of the terms of use applies to Standard Services 'including where you have changed your Account setting under a Coding Harness Subscription' (then MGPT, Meta as processor). In the preset Contributor mode, the Contributor terms apply (section 6, no personal data).

yes: For Standard Services, the terms of use (section 5.1) incorporate the 'Meta Global Processor Terms' (MGPT, as of 2026-03-20) and the data security terms; Meta acts as processor. For European data, Europe-specific MGPT terms and an addendum on global data transfer apply.

no (the processor terms are incorporated in section 5 only for Standard Services; terms of use 6.2: no sensitive, confidential or personal information may be submitted to discounted services)

Privacy policy (as of 2026-10-11) and Europe Addendum (effective 2026-08-24): User Content and feedback are used for training and improving the models on the basis of legitimate interests; objection possible in the settings (reference to the Consumer FAQ). Security FAQ: 'improve the model' toggle in the data controls. Training summary Grok 4.7: interactions with Grok may be used for training 'subject to privacy settings, controls, user requests, opt-outs'.

Privacy policy (as of 2026-10-11) and Europe Addendum (effective 2026-08-24): User Content and feedback are used for training and improving the models on the basis of legitimate interests; objection possible in the settings (reference to the Consumer FAQ). Security FAQ: 'improve the model' toggle in the data controls. Training summary Grok 4.7: interactions with Grok may be used for training 'subject to privacy settings, controls, user requests, opt-outs'.

Enterprise terms (as of 2026-08-14): SpaceXSI does not use User Content to train foundation models, LLMs or other AI systems, 'subject to disclosures to Customer and Customer-controlled user settings'; SpaceXSI may use de-identified or aggregated data except with ZDR. Business guide (as of 2026-10-10): team workspaces with 'Privacy guarantees as outlined in SpaceXSI's terms of service' (link to the Enterprise terms). Enterprise FAQ: 'we do not use your business data, including inputs (prompts) or outputs (answers), for training our models'; in some cases SpaceXSI offers free credits in exchange for permission to train.

Enterprise terms (as of 2026-08-14): SpaceXSI does not use User Content to train foundation models, LLMs or other AI systems, 'subject to disclosures to Customer and Customer-controlled user settings'; SpaceXSI may use de-identified or aggregated data except with ZDR. Business guide: team workspaces with 'Privacy guarantees as outlined in SpaceXSI's terms of service'. Enterprise FAQ: 'we do not use your business data, including inputs (prompts) or outputs (answers), for training our models'; in some cases SpaceXSI offers free credits in exchange for permission to train.

API documentation (Security FAQ): 'SpaceXSI never trains on your API inputs or outputs without your explicit permission.' Enterprise terms (as of 2026-08-14): SpaceXSI does not use User Content to train foundation models, LLMs or other AI systems, 'subject to disclosures to Customer and Customer-controlled user settings'. Except with ZDR, SpaceXSI may use de-identified or aggregated data for any lawful purposes.

API documentation (Security FAQ): 'SpaceXSI never trains on your API inputs or outputs without your explicit permission.' Enterprise terms: no training on User Content; with ZDR also no creation of de-identified data. With ZDR the 30-day storage does not apply; inputs and outputs are not written to disk. The documentation points out that ZDR is not recommended for most customers.

Privacy Center 'How Meta uses information for generative AI models' (translated) (q9): training sources include 'interactions with features that are part of AI at Meta' (translated), public information in Meta products, internet content and licensed data. Private messages to friends and family are not used unless they are knowingly shared with the AIs. There is a right to object to the use of interactions with AI at Meta and of public information (via the account or regional forms according to the Privacy Center; applies to future training). In the UK and the EU, no public information from accounts of under-18s.

Subscription docs: Muse Code subscriptions are in Contributor mode by default ('default to Contributor mode'); content is then used to train future Meta models. The mode can be switched off in the settings at any time; thereafter, future content is not used for training.

Terms of use (as of 2026-10-09, 5.1): Meta does not use content from Standard Services to train Meta models. Pricing page: 'In Standard mode, you pay standard prices and your content is not used to train Meta models.' Data commitments: content from paid services is not used for training; customer data (account, roles, profile fields) in no case.

Pricing page: Contributor mode 'at heavily discounted token prices, in exchange for permission to use your content to train future Meta models'. Terms of use (section 6): content of discounted services may be used by Meta for training.

Europe Addendum (13.3): 'We process all of your personal information in the United States, where we maintain our primary data centers.'

Europe Addendum (13.3): 'We process all of your personal information in the United States, where we maintain our primary data centers.'

The privacy policy does not apply to data processed for business customers. DPA (q10): processing 'on a global basis outside Europe', including in the USA; sub-processors in the US, one subsidiary in the UK. The pricing page mentions data residency only in the Enterprise section; EU region not documented.

The privacy policy does not apply to data processed for business customers. DPA (q10): processing 'on a global basis outside Europe', including in the USA; sub-processors in the US, one subsidiary in the UK. Pricing page (Enterprise): 'Data residency: Control where your data lives' without a region specified; EU region not documented.

Regions documentation (q3): the standard endpoint api.x.ai does not guarantee a processing region; requests may be routed between regions. Optional US endpoint us.api.x.ai: request handling, inference, moderation and stored request data in the USA (only grok-4.7 and grok-4.6, without image, video and voice APIs; Files, Collections and server-side tools may be processed outside the USA; token prices +10 %). DPA (q10): processing 'on a global basis outside Europe', including in the USA. Sub-processors in the US, one subsidiary in the UK. An EU region is not documented.

As API: global endpoint without a region guarantee, US endpoint optional (ZDR also applies there). No EU region documented.

not publicly documented on the retrieved pages.

Region: not publicly documented. According to the docs, subscription features and availability may vary by region.

Region and EU option for the API: not publicly documented. Terms of use (section 6): 'Meta does not receive, determine, or route traffic based on location of your End Users'. The sub-processor list linked in section 5 names data centre operators predominantly in the USA as well as in Denmark, Sweden and Ireland; which of them are used for the Model API is not stated. Model docs: 'Model availability can vary by country' (reference to the Geographic Use Policy).

Region and EU option: not publicly documented.

• Europe Addendum (q8): service providers as per the sub-processor page (q11, see API plan); Brave Software, Inc. receives parts of the inputs for search results; payment processing by Google Payments, Apple Pay and Stripe

• Europe Addendum (q8): service providers as per the sub-processor page (q11, see API plan); Brave Software, Inc. receives parts of the inputs for search results; payment processing by Google Payments, Apple Pay and Stripe

• According to the sub-processor list (q11): AWS, Google Cloud, Oracle (cloud infrastructure), Vercel (web hosting), MongoDB, Inc. (cloud database), Cloudflare and Hive (content delivery), LiveKit (real-time media), Axiom, Chronosphere, ClickHouse, Elasticsearch, GrowthBook, Hubspot, Okta, Twilio, Wiz, WorkOS, Zendesk (services and support), Sentry (error monitoring), Stripe and PayPal (Braintree) (payments), SpaceXSI affiliates (cloud infrastructure), each located in the US
• SpaceXSI subsidiary (services and support): located in the UK (q11)

• According to the sub-processor list (q11): AWS, Google Cloud, Oracle (cloud infrastructure), Vercel (web hosting), MongoDB, Inc. (cloud database), Cloudflare and Hive (content delivery), LiveKit (real-time media), Axiom, Chronosphere, ClickHouse, Elasticsearch, GrowthBook, Hubspot, Okta, Twilio, Wiz, WorkOS, Zendesk (services and support), Sentry (error monitoring), Stripe and PayPal (Braintree) (payments), SpaceXSI affiliates (cloud infrastructure), each located in the US
• SpaceXSI subsidiary (services and support): located in the UK (q11)

• According to the sub-processor list (q11): AWS, Google Cloud, Oracle (cloud infrastructure), Vercel (web hosting), MongoDB, Inc. (cloud database), Cloudflare and Hive (content delivery), LiveKit (real-time media), Axiom, Chronosphere, ClickHouse, Elasticsearch, GrowthBook, Hubspot, Okta, Twilio, Wiz, WorkOS, Zendesk (services and support), Sentry (error monitoring), Stripe and PayPal (Braintree) (payments), SpaceXSI affiliates (cloud infrastructure), each located in the US
• SpaceXSI subsidiary (services and support): located in the UK (q11)

• According to the sub-processor list (q11): AWS, Google Cloud, Oracle (cloud infrastructure), Vercel (web hosting), MongoDB, Inc. (cloud database), Cloudflare and Hive (content delivery), LiveKit (real-time media), Axiom, Chronosphere, ClickHouse, Elasticsearch, GrowthBook, Hubspot, Okta, Twilio, Wiz, WorkOS, Zendesk (services and support), Sentry (error monitoring), Stripe and PayPal (Braintree) (payments), SpaceXSI affiliates (cloud infrastructure), each located in the US
• SpaceXSI subsidiary (services and support): located in the UK (q11)

• According to the list linked in the terms of use (title 'Meta Horizon Managed Solutions – List of Sub-processors' (translated), valid from 2023-11-13): data centre infrastructure by Andale, Greater Kudu, Goldframe, Morning Hornet, Offprints, Omanyte, Paile, Raven Northbrook, Scout Development, Siculus, Sidecat, Stadion, Starbelt, Vitesse, Winner (d/b/a Ernst), Woolhawk (each USA), Cassin Networks (Denmark), Pinnacle Sweden (Sweden), Runways Information Services (Ireland)
• Meta subsidiaries according to the same list: Meta Platforms, Inc. and Meta Operations, LLC (USA), Meta Platforms Ireland Limited (Ireland), Meta Canada, Meta France, Meta Germany, Meta Italy, Meta Netherlands, Meta Poland, Meta Spain, Meta UK, Meta Singapore, Meta Platforms International Operations Philippines

0 USD (pricing page)

Plus 8 USD; Premium 30 USD; Super 100 USD; Ultra 200 USD per month, each excl. taxes; subscriptions are sold via X Subscription Corp. and include X features (pricing page). The Consumer FAQ still names 'SuperGrok' and 'SuperGrok Heavy'.

from 30 USD per user per month ('$30+/user/month'; seat types combinable), excl. taxes (pricing page)

Custom (pricing page: 'Custom', contact sales; payment by invoice/bank transfer only in the Enterprise plan)

Per 1M tokens, in USD, below / from 200k prompt tokens (pricing page): grok-4.7 and grok-4.6 input 2.00 USD / 4.00 USD, cache 0.50 USD / 1.00 USD, output 6.00 USD / 12.00 USD; grok-4.5 input 2.00 USD / 4.00 USD, cache 0.30 USD / 0.60 USD, output 6.00 USD / 12.00 USD; grok-4.3 and grok-4.20 variants input 1.25 USD / 2.50 USD, cache 0.20 USD / 0.40 USD, output 2.50 USD / 5.00 USD; grok-build-0.1 input 1.00 USD / 2.00 USD, cache 0.20 USD / 0.40 USD, output 2.00 USD / 4.00 USD. Tools: web search, code execution, file search 5 USD per 1,000 calls each; X search 5 USD per 1,000 posts or 10 USD per 1,000 profiles; Collections search 2.50 USD per 1,000 calls. US endpoint 10 % surcharge.

As API (ZDR surcharge not stated in the documentation)

not publicly documented on the retrieved pages

Flat monthly price in USD (Help Center 'What is a Muse Code subscription?'): Everyday Usage USD 5/month, High Usage USD 15/month (5x usage), Power Usage USD 50/month (20x), Maximum Usage USD 200/month (80x); automatic monthly renewal; not stated whether VAT is included. According to the docs, benefits and availability may vary by region.

Per 1M tokens, in USD (pricing page): Muse Spark input USD 1.25, cached input USD 0.15, output USD 4.25; Muse Wonder input USD 4.00, cached input USD 0.20, output USD 20.00; no surcharge for long context. Muse Image USD 0.01 per image; Muse Voice Transcribe USD 0.18 per hour (ZDR at the same price); web search grounding USD 2.50 per 1,000 search queries; SAM 3.1 USD 2.50 per 1,000 images, USD 0.20 per 1,000 video frames; no minimum commitment. Taxes are added (terms of use 13.1: the customer bears applicable sales, value added or similar taxes).

Per 1M tokens, in USD (pricing page): input USD 0.10, cached input USD 0.002, output USD 0.20. Taxes are added (terms of use 13.1: the customer bears applicable sales, value added or similar taxes).

yes (limited chat, agent, image and Build usage according to the pricing page)

yes (from Plus: Grok Chat, Voice, Imagine, Build, API access and Cursor; from Premium: Grok Bot, latest models, video and image generation, 4x usage compared with Plus; Super/Ultra with 15x and 50x usage respectively compared with Plus)

yes (according to the Teams comparison table: Grok Chat, Imagine, video generation, Build, API access, Grok Bot, Cursor agent, pooled usage, team management, central billing, team-wide privacy settings, custom chat retention, SAML/OIDC SSO, combinable seat types; self-service by card. Business guide: connectors to Google Drive, Outlook, SharePoint, Teams, Salesforce, among others)

yes (as Teams; additionally, according to the pricing page, SCIM, audit logs and service accounts, spend controls, group settings, purchase on invoice, custom rate limits, dedicated infrastructure, 'Data residency', among others)

yes (models grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20, grok-build-0.1; tools such as web and X search, code execution; image, video and voice APIs)

yes (as API; without functions that require storage: Files, Collections, Batch API, stateful Responses API, among others)

yes (Meta AI assistant in Meta products and on meta.ai; part of 'AI at Meta' (translated) according to the Privacy Center)

yes (coding agent Muse Code with Muse Spark, Muse Wonder and Muse Image; Power and Maximum additionally Muse Wonder Full; use only via the Muse Code CLI)

yes (models muse-wonder-1.0 ('in preview' according to the model docs), muse-spark-1.3, -1.2, -1.1; image model Muse Image, Muse Voice Transcribe, SAM 3.1; web search grounding)

yes (muse-spark-1.3-contributor, muse-spark-1.2-contributor; not for Muse Wonder and Muse Voice Transcribe)