SpaceXAI's new Grok terms make "indirect competitor" a deployment risk

SpaceXAI's revised acceptable use policy, listed as effective August 14, 2026, bars customers from using its services or any output to develop, or to assist anyone in developing, machine-learning models or any products or services that compete with SpaceXAI, "directly or indirectly" ^[1]. Violations can end in account suspension or termination, and the rules apply to consumers, developers and businesses alike ^[2], which moves the definition of "competitor" out of the legal annex and into your architecture diagram.

The predecessor text, dated June 26, 2026, was narrower: it stopped customers using outputs to develop machine-learning models or related AI services that competed with xAI ^[3]. Two edits do the work, on RuntimeWire's reading. "Assist anyone" extends the restriction beyond the account holder's own product ^[4]. Swapping "related AI services" for "any products or services" widens the enforcement field, and the policy does not confine competition to foundation models ^[5].

Scope is the problem. SpaceXAI's company page pitches scientific discovery, reasoning, voice and generative media ^[6], and RuntimeWire notes that Grok now spans chat, coding, voice, image, video and agent products, so a customer product can sit several layers above the API and still overlap ^[7]. The categories it flags as exposed are the ones most teams are actually building: model routers, coding agents, synthetic-data pipelines, evaluation tools, and applications that switch between frontier models ^[8]. A router that can fail over to another vendor is, by construction, hard to distinguish from an indirect competitor.

The same policy prohibits scraping, harvesting, reselling and distilling inputs or outputs, reverse engineering, removing provenance metadata or watermarks, evading rate limits, prompt injection against the service, and access through unauthorized bots or scripts ^[9]. Most of that is standard for a frontier vendor. The distillation and metadata clauses are the ones that will surprise teams running comparative benchmarks.

The paperwork itself is not settled. An indexed version of xAI's official page still lists June 26, 2026 as the effective date, and the page RuntimeWire examined links to that June version as its predecessor ^{[10] [11]} - 49 days between the two dates ^[12]. Naming is equally mixed: SpaceXAI appears in the heading and throughout the new rules, while the footer still identifies X.AI LLC and the linked consumer terms describe an agreement with X.AI LLC ^[13]. That follows SpaceX's February 2 acquisition of xAI, which put the models, X distribution and SpaceX infrastructure under one parent ^[14].

The commercial push runs in the other direction. RuntimeWire reported on August 12 that SpaceXAI shipped Grok 4.6 with a 500,000-token context window and integrations with Cursor and Grok Build ^[15], and the company's homepage claims a unified API handling more than 1 million calls per day across at least five model families - SpaceXAI's own figures, with no disclosure of how many developers or paying businesses sit behind them ^[16]. In July, selected builders at a San Francisco Grokathon were given access to Grok models and X APIs ^[17]. xAI announced a $20 billion Series E on January 6 ^[18]. The invitation to build is wide; the licence to keep what you build is narrower than it was in June.

Watch which text actually governs when the two effective dates are reconciled, and watch for the first named suspension. Until a vendor shows how it reads "indirect," the safe assumption for anyone with a second model provider is that the clause reads however the vendor needs it to.