Product and intended use
Define the configuration, people, information and tasks being assessed.
What is sold and under which terms
Public claim only
The vendor's product navigation (read 2026-09-27) lists Grok, Build, Imagine, Voice, Bot and Grokipedia, with tiers published as Business (x.ai/grok/business) and Government (x.ai/grok/government). 'Legal' appears only under Solutions, alongside Customer Support and Security; there is no separate legal product page, legal SKU or legal pricing tier. The Enterprise Terms state they are 'for enterprise (business) users of the SpaceXAI API and related SpaceXAI Services (including Grok) who are at least 18 years old' and that consumer use of Grok is governed by a separate Terms of Service - Consumer.
Ask: Which SKU and which terms would apply to our firm: the Business tier, an enterprise Order Form, or API access, and is the Zero Data Retention API included in that SKU at no extra charge?
What the vendor says it does for legal teams
Public claim only
The legal solutions page (read 2026-09-27) is headed 'AI for legal teams' and names five capabilities: contract analysis ('extract key terms, flag non-standard clauses, and compare against your templates in seconds'); legal research ('Search across case law, statutes, and regulatory guidance with natural language queries. Get structured summaries with citations'); compliance monitoring ('Track regulatory changes across jurisdictions and map them to your internal policies'); document drafting ('Generate first drafts of NDAs, amendments, and policy documents based on your existing templates and guidelines'); and due diligence ('Process large document sets for M&A, audits, and investigations'). Three linked use-case pages cover document processing, research synthesis and long-form writing. The page describes use cases for the vendor's general models; no legal-specific retrieval source or legal corpus is named.
Ask: Which of the five legal capabilities are product features with their own interface, and which are prompting patterns against the general model with documents we supply?
Jurisdictional coverage of the legal use cases
Not confirmed
The legal solutions page states that Grok 'understands legal language, citation formats, and jurisdictional nuance' and offers to 'Track regulatory changes across jurisdictions', but names no jurisdiction, no Australian court, tribunal or legislation source, and no currency date. No page read publishes a jurisdiction list or a legal content index.
Ask: For Australian federal and state law, what sources does Grok draw on when answering a legal research question, and to what date is each current?
How the service is delivered
Public claim only
The Enterprise Terms (14 August 2026) grant a subscription right to access the Services identified in an Order Form and to use the APIs to build integrations ('Bundled Services') that may be distributed to the customer's own end users, with no direct contractual relationship between the vendor and those end users. The vendor also offers a Zero Data Retention-enabled API as a distinct delivery option. Neither the Enterprise Terms nor the DPA describes an on-premises, private-cloud or self-hosted deployment, and the vendor's security page was not read within the credit budget.
Ask: Is any deployment other than the multi-tenant hosted service and the ZDR-enabled API available to us, and can inference be pinned to a named region?
Beta and preview functionality
Public claim only
Enterprise Terms clause 1.4 states that beta offerings are conspicuously identified as beta, are not part of the Services, are optional and at the customer's sole risk, are provided 'AS-IS' and 'AS-AVAILABLE' with no warranties, 'may contain errors, defects, bugs, or inaccuracies that could cause failures, data corruption, loss of data or information', and that the vendor disclaims all liability arising from their use. The performance warranty expressly does not apply to any services provided on a no-charge, trial, beta, preview or evaluation basis.
Ask: Which features we would rely on are currently beta offerings under clause 1.4, and how are they labelled inside the product?
Check for vendor content directing how AI describes the product
Not confirmed
A deliberate check was made on 2026-09-27. A site map of x.ai returning 71 URLs shows no /llms.txt, /ai-info, /ai, /for-ai or equivalent path. The legal solutions page, the Enterprise Terms, the DPA and the enterprise FAQ, all read in full, contain no block addressed to AI models, assistants, LLMs or crawlers and no 'Instructions for AI Assistants' heading. The enterprise FAQ is in question-and-answer form and contains no comparative or superlative answer about the vendor's own products. Not checked within the credit budget: x.ai/legal/brand-guidelines, x.ai/legal/faq (consumer) and x.ai/security.
Ask: Does the vendor publish any page, file or embedded block that tells AI models, assistants or crawlers how to describe Grok, including on the brand guidelines page?
Data, controls and requirements
Compare the actual configuration with firm, client and jurisdiction requirements, supported by evidence.
Authority to put client material into the service
Public claim only
Enterprise Terms clause 3.1 requires the customer to represent and warrant on a continuing basis that it has and will maintain all rights, licences, consents, permissions and authorisations necessary to submit input and to grant the vendor the rights in clause 3.2. Clause 11.1 makes the customer solely responsible for determining the legality of submitting personal data, for obtaining all necessary consents, rights and authorisations, and for any consequences of those submissions. The DPA clause 2.3 requires the customer to ensure its processing instructions comply with applicable data protection laws, and clause 2.4.5 states the vendor is not responsible for monitoring the customer's compliance or assessing whether its instructions are lawful. The DPA also states the vendor 'requires that Customer does not submit Personal Data to SpaceXAI's non-ZDR Services'.
Ask: Given the DPA requires personal data to be processed only through the ZDR-enabled API, how do we prevent client personal information reaching the non-ZDR web or mobile apps, and what does the vendor do if it happens?
Which model answers, and whose it is
Not confirmed
No model name, version or third-party model provider is named on the legal solutions page, in the Enterprise Terms, in the DPA or in the enterprise FAQ. The Enterprise Terms define 'SpaceXAI Technology' to include the vendor's own 'models, algorithms, data, or other materials', and the DPA names SpaceXAI LLC as the sole data importer. The model documentation at docs.x.ai/developers/models was not read within the five-operation credit budget, and the vendor's news index carries unread pages whose URL slugs refer to model releases and to availability on third-party clouds; nothing is asserted from those slugs.
Ask: Which Grok model and version serves the legal use cases on x.ai/solutions/legal, is any third-party model or third-party cloud inference involved, and are we notified when the serving model changes?
Subprocessors and how changes are notified
Public claim only
DPA clause 3 (effective 22 September 2026) records a general written authorisation for the vendor to engage subprocessors, publishes the list at x.ai/legal/subprocessor-list, requires written agreements with data protection obligations substantially the same as the DPA, and keeps the vendor liable for subprocessor acts. Changes are notified at least 15 days before taking effect, with a 15-day written objection right on reasonable data protection grounds; if no resolution is reached the vendor may either not appoint the subprocessor or allow termination of the affected part of the Services, and that termination right is the customer's sole remedy. Silence is deemed approval. The subprocessor list itself was not read within the credit budget.
Ask: Which subprocessors process inputs and outputs for the business tier, in which countries do they operate, and can we receive subprocessor change notices by email to a nominated address?
Training on customer content
Public claim only
Enterprise Terms clause 3.1 states the vendor 'will not use any User Content to train any foundation models, large language models, or other artificial intelligence systems or to develop any new products, services, or features, subject to disclosures to Customer and Customer-controlled user settings.' The enterprise FAQ states 'No, we do not use your business data, including inputs (prompts) or outputs (answers), for training our models' and adds 'In some instances we may offer free credits in exchange for you agreeing to permit us to train on your business data.' Separately, clause 3.3 allows the vendor to create and use de-identified and aggregated data derived from the customer's use for any lawful purpose, and to own it, except where the customer elects the ZDR API.
Ask: Which user-controlled settings could turn training on, who in our tenancy can change them, and will you confirm in the Order Form that no free-credit training arrangement applies to us?
Feedback, ratings and derived data
Public claim only
Enterprise Terms clause 4.2 assigns to the vendor all right, title and interest in any suggestions, ideas, improvements, recommendations and other feedback relating to the Services, including intellectual property rights, and grants a perpetual, irrevocable, worldwide, royalty-free, sublicensable licence to exploit that feedback for any purpose without restriction, accounting, compensation, credit or attribution. Clause 4.1 includes 'feedback, or suggestions' within the vendor-owned SpaceXAI Technology. Clause 3.3 permits the vendor to own and use de-identified and aggregated data derived from the customer's use of the Services for any lawful purpose unless the ZDR API is used.
Ask: If a lawyer pastes a clause or a matter fact into a feedback or bug report, does clause 4.2 give you rights over that text, and can feedback submission be disabled for our tenancy?
How long content is kept
Public claim only
Enterprise Terms clause 3.4 sets automatic permanent deletion of all user content no later than 30 days after the end of the interaction or session, with four stated exceptions: a different period agreed in an Order Form, Documentation, conspicuous in-product notice or customer-selected self-service setting; a legal requirement; or where reasonably necessary for safety, security, compliance, moderation, abuse prevention or investigation, in which case only the minimum data is kept. Under the ZDR API, content exists only transiently and is deleted on the earlier of one hour after the inference request completes or delivery of the response, with no logs, backups or persistent copies. The DPA security appendix states that following a deletion request, login data, prompt and response pairs and billing information are deleted 'typically within 30 days'.
Ask: For our tenancy, what is the configured retention period, where is it recorded, and what triggers the safety and abuse exception that allows longer retention?
Deletion on request and at exit
Public claim only
DPA clause 9 states that on termination of the agreement the vendor will delete personal data in its possession, except where required to retain copies under applicable law, in which case the data is isolated and protected from further processing; the vendor may continue to process information the customer instructed it to de-identify, anonymise or aggregate. The DPA security appendix states 'Customers are able to delete their data at will' and that after a deletion request systems automatically delete all login data, prompt and response pairs and billing information 'typically within 30 days'. Under the ZDR API the vendor states deletion is immediate after processing and that deleted content cannot be recovered or produced, including in response to subpoenas, regulatory requests or legal process.
Ask: What self-service deletion controls exist for an administrator, what evidence of deletion is provided, and how do backups age out after a deletion request?
Vendor personnel access to content
Public claim only
The enterprise FAQ states the vendor 'may use automated content classifiers and safety tools' to understand how the services are used and to check compliance with its Terms and Acceptable Use Policy, and that 'A limited number of authorized SpaceXAI personnel may review your data when legally required, such as investigating security incidents and potential misuse of our services.' DPA appendix 2 states access to user content is restricted to authorised personnel subject to confidentiality agreements and background checks where legally permissible, that access is de-identified using per-customer hashed identifiers with data masking, and that access is monitored and alerted on anomalous access. DPA clause 4.2 states the vendor does not intend to disclose personal data to law enforcement or government authorities unless instructed by the customer or as necessary to comply with law or a binding order, and will disclose only the minimum information required.
Ask: In the last 12 months, how many times did personnel access customer content for safety or incident investigation, and are customers notified when their content is reviewed?
Authentication and administrative control
Public claim only
DPA appendix 2 states access to the platform is restricted by authentication and authorisation policies implemented in the vendor's software stack, that identification is performed by verifying the email and domain of the party accessing the platform, that authorisation checks then determine the level of access and the workspaces available, and that 'Workspace administrators (customers) are able to authorize additional parties as they see fit.' It also records least-privilege and need-to-know technical controls, unique identities, password complexity enforcement, revocation on termination or role change, and regular access reviews. Enterprise Terms clause 1.2 makes the customer liable for permitted users and requires credentials to be kept confidential.
Ask: Does the business tier support single sign-on with our identity provider, enforced multi-factor authentication and SCIM de-provisioning, and can an administrator see per-user activity logs?
Encryption in transit and at rest
Public claim only
DPA appendix 2 states all user content is de-identified using per-customer hashed identifiers and encrypted at rest and in transit using AES-256 for data at rest and TLS 1.3 in transit, that a per-customer decryption key is used to retrieve data before transmission over a TLS 1.3 channel, and that user content at rest is encrypted with per-customer AES-256 keys 'handled only by machine-based services, and not made available to any employees'. Enterprise Terms clause 11.3 commits to commercially reasonable physical, technical, organisational and administrative measures as described at x.ai/security and appendix 2 of the DPA.
Ask: Are the per-customer AES-256 keys customer-managed or vendor-managed, and is a customer-managed key or bring-your-own-key option available on the business tier?
Certifications and audit reports
Public claim only
No certification is named in any document read. DPA clause 5.2 states that on written request, and subject to reasonable notice and confidentiality agreements, the vendor will provide access to documentation demonstrating compliance with the DPA 'including copies of certifications, audit reports, and/or other relevant documentation', and that at its discretion it may instead provide a summary of third-party certification or audit results. DPA appendix 2 states the vendor 'has adopted the NIST 800-171 Rev.3 framework as a baseline for our internal security standards', has a written information security policy and a dedicated security team, and that third-party penetration testing is conducted yearly. No SOC 2, ISO/IEC 27001, ISO/IEC 42001 or IRAP status is claimed; the vendor's security page at x.ai/security was not read within the credit budget.
Ask: Which third-party attestations do you hold today for the business tier, will you provide the SOC 2 Type II report or ISO certificate under NDA, and has any IRAP assessment been done?
Contracting entity, governing law and liability
Public claim only
The Enterprise Terms (last updated 14 August 2026) are entered into with SpaceXAI LLC, are governed by the laws of the State of Texas without regard to conflict-of-laws principles, and give the state and federal courts of Wichita County, Texas exclusive jurisdiction; clause 13.4 is a class action and jury trial waiver. Each party's total cumulative liability is capped at the total fees paid and payable in the 12 months immediately preceding the event giving rise to the first claim, with stated exceptions. The vendor gives an intellectual-property indemnity for authorised use of the Services, subject to listed exclusions including use with unapproved third-party products and bundled services. No Australian entity, Australian governing law option or Australian Consumer Law carve-out appears in the Enterprise Terms or the DPA.
Ask: Will you contract through an Australian entity or accept Australian governing law and the non-excludable Australian Consumer Law guarantees, and what are the carve-outs to the 12-month fee liability cap?
Cross-border processing
Public claim only
DPA clause 8.1 states the customer acknowledges and agrees that the vendor 'may transfer and process Personal Data on a global basis outside Europe as necessary to provide the Services, including in the United States and other countries where SpaceXAI and its Subprocessors maintain data processing operations.' Clause 8.2 incorporates the EU standard contractual clauses for EU, UK and Swiss restricted transfers, with the clauses governed by the law of Ireland for EU transfers. No region-selection, data-residency or in-country processing option is offered in the DPA or the Enterprise Terms, and no Australian transfer mechanism or Australian Privacy Principle 8 undertaking appears. DPA appendix 2 states 'Multiple geographic zones are in place to ensure service availability is not impacted by a single point of failure' without naming them.
Ask: In which countries will our inputs and outputs actually be processed and stored, and is any region pinning or Australian data residency available for the business tier?
Security incident notification
Public claim only
DPA clause 5.3 commits the vendor to notify the customer without undue delay and, where feasible, no later than 48 hours after becoming aware of a security incident, describing its nature and other information the vendor can reasonably disclose. Clause 5.4 lists the content of the notification, including categories and numbers of individuals and records affected, likely consequences, and measures taken, and states the customer 'has the right to control the breach notification process' including notifying individuals and regulators unless the law requires otherwise. Unsuccessful log-in attempts, authorised penetration testing, pings, port scans and denial-of-service attacks are excluded from the definition of a security incident. Notification is stated not to be an acknowledgement of fault or liability.
Ask: Who is notified at our organisation within the 48 hours, and how does the notification work in practice when ZDR means you hold no copy of the affected content?
How terms, policies and the service change
Public claim only
Enterprise Terms clause 13.10 states no modification or amendment is effective unless made in writing and executed by a duly authorised representative of each party. However, the Acceptable Use Policy is incorporated 'as it may be updated from time to time', and the Documentation at docs.x.ai is incorporated 'as updated from time to time'. DPA clause 10.4 states the vendor may modify or update the DPA in its sole discretion provided the level of data protection is not decreased, will update the effective date and notify customers of material changes, and that continued use constitutes acceptance. The terms page lists seven previous versions from 5 September 2024 to 12 May 2026, and the DPA links one previous version dated 24 August 2026.
Ask: How and to whom are changes to the Acceptable Use Policy, the Documentation and the DPA notified, and how much notice do we get before a change that affects model behaviour or retention?
Termination and getting content out
Public claim only
Enterprise Terms clause 6.2 allows either party to terminate on 30 days' notice at its sole discretion, with a pro-rata refund of prepaid unused fees where the vendor terminates; a party may also terminate for uncured material breach after 30 days, cessation of business or insolvency. On termination, all Order Forms terminate, the customer must cease use, unused prepaid credits are forfeited and unpaid minimum commitments remain payable. Confidentiality obligations survive five years, and trade secret obligations survive indefinitely. Neither document read contains an export, migration-assistance or data-portability obligation; under ZDR the vendor states it holds no content to return.
Ask: On exit, in what format and over what period can we export prompt and output history and workspace configuration, and is any transition assistance available?
Third-party services and integrations
Public claim only
Enterprise Terms clause 3.5 states the Services may allow the customer, at its option, to access, use or integrate with third-party products or services, which are governed by the customer's terms with that provider, are provided 'AS-IS' with no vendor warranty, and for which the vendor accepts no liability including for security or results; the vendor provides no support for them. Clause 3.2 states that if the customer directs or configures the Services to transmit input or output to any third-party platform, the customer represents it has the necessary rights and remains fully responsible. No connector catalogue, and no integration with any document management or practice management system, is named in the documents read.
Ask: Which connectors to document management, email and practice management systems exist for the business tier, and what permissions does each request in our tenancy?
Whether answers draw on live web content
Not confirmed
Nothing in the legal solutions page, the Enterprise Terms, the DPA or the enterprise FAQ states whether Grok retrieves live web content when answering, or whether such retrieval can be disabled. The legal solutions page refers only to searching 'across case law, statutes, and regulatory guidance' without naming a source or saying whether the source is the open web. No setting, control or logging of external retrieval is described in any document read.
Ask: When Grok answers a legal question, does it retrieve live web pages, can an administrator disable that retrieval, and are the retrieved URLs logged for each answer?
Autonomous actions and agents
Not confirmed
No document read describes autonomous or agentic action, tool use, or the ability of the product to act in external systems on the legal use cases; the legal solutions page describes generation and analysis only. The vendor's site map lists a 'Grok Bot' product with a marketplace of named bots and guides, and a 'Build' product, none of which were read within the five-operation credit budget, so no capability, permission model or control is asserted from them.
Ask: Can Grok or Grok Bot take actions in connected systems, such as sending email or writing to a document store, and what administrative approval sits in front of each action?
Verification of output before use
Public claim only
Enterprise Terms clause 3.2 states that the customer acknowledges outputs are probabilistic and may not be unique, 'may be incomplete, inaccurate, biased, objectionable, inappropriate, offensive, contain "hallucinations," or otherwise unsuited to Customer's purpose', and that the customer 'is solely responsible for independently evaluating the accuracy, completeness, and suitability of any Output before relying on or distributing it (including through human review where appropriate)'. The same clause states the vendor has no liability for damages arising from any output. The Services are otherwise provided 'AS IS' and 'AS AVAILABLE' except for a performance warranty that the Services will perform materially in accordance with the Documentation.
Ask: What in-product features support verification of a legal answer, such as citation links to the source text or a confidence indicator, and can outputs be blocked from export until a reviewer signs off?
