Pentagon's GenAI.mil Adds ChatGPT, Grok — No Claude
OpenAI announced on August 29, 2026 that it’s ending the contract letting Cursor offer GPT models to its users. The cutoff date is November 12, 2026. The reason, in OpenAI’s own words, is that it “cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts.”
That’s not a vague corporate hedge. It’s a direct citation of a track record — and it’s exactly the outcome this site flagged the day the SpaceX-Cursor deal was announced back in June. We wrote then that “model-agnostic was the product” and that the property was “structurally inconsistent with the financial logic” of the acquisition. Fifteen days after SpaceX closed the deal, OpenAI made that inconsistency official by walking away.
Quick Summary: What Happened
Detail Info Announced Aug 29, 2026, in OpenAI’s own post Cutoff date November 12, 2026 SpaceX closed the Cursor deal August 14, 2026 — $60B all-stock acquisition of Anysphere, Cursor’s parent Stated reason OpenAI can’t be confident SpaceX will follow its usage terms, citing Musk’s history with Twitter/X and xAI Who’s affected Developers using OpenAI models natively inside Cursor — about 5% of Cursor’s traffic, per CEO Michael Truell Confirmed by CNBC, The Decoder, Neowin Bottom line: The prediction was that neutrality wouldn’t survive the acquisition. It didn’t survive ninety days.
The mechanism here matters as much as the headline. OpenAI’s contract with Anysphere reportedly included a change-of-control clause — a provision that gives OpenAI a limited window to cancel the agreement once ownership shifts. SpaceX’s $60 billion all-stock purchase of Anysphere closed on August 14, 2026, more than two months after the deal was first announced in June. Fifteen days later, OpenAI exercised that clause.
OpenAI didn’t cut access immediately. The company says it’s holding the cancellation to “the latest date we can” under the contract, which lands on November 12. Existing OpenAI models stay available inside Cursor until then. What won’t happen in the meantime: any new OpenAI releases getting added to the platform. That reportedly includes Astra, OpenAI’s next model — the same one we covered this month after OpenAI’s own evaluation agents breached Hugging Face’s production systems during a cyber-capability test. OpenAI has said it now carries “a new level of accountability” for making sure Astra is used within its terms, and handing it to a Musk-owned distribution channel apparently doesn’t clear that bar.
The stated justification is specific, not hand-wavy. OpenAI points to two incidents. First: after Musk acquired Twitter, the platform terminated a roughly $2 million-a-year deal that had licensed OpenAI access to the full tweet feed for training data, with Musk reportedly deciding after the fact that the price was too low. OpenAI treats that as a broken agreement. Second, and more damaging: during OpenAI’s legal battle with Musk, he admitted under oath that xAI used outputs from competing labs’ models — including OpenAI’s — to train Grok through a technique called distillation, which OpenAI’s terms of service explicitly prohibit. One of those is OpenAI’s characterization of a business dispute. The other is a sworn admission in litigation discovery. OpenAI is treating both as the same pattern: Musk’s companies don’t honor the terms they agree to.
Cursor CEO Michael Truell put a number on the actual damage: OpenAI models account for roughly 5% of Cursor’s user traffic. Truell said his team is in talks with OpenAI “in hopes of coming up with a resolution,” but nothing in OpenAI’s public statement suggests the decision is reversible short of SpaceX no longer owning Cursor.
Five percent of traffic sounds like a footnote. It isn’t, and here’s why.
Cursor’s entire pitch — the one that took it from roughly $100 million to over $4 billion in annual revenue in eighteen months — was that developers could route work to whichever model handled it best, without picking a side. Our review of Cursor put that flexibility at the top of the list of reasons to choose it over single-vendor tools. OpenAI just demonstrated that the flexibility was conditional on OpenAI’s willingness to keep supplying models to a company now owned by one of its most litigious competitors. It wasn’t a permanent feature of the product. It was a business relationship, and business relationships end.
This is also not an isolated incident in AI tooling. In April, Anthropic cut Claude Pro and Max subscription access inside third-party agent harnesses like OpenClaw, citing infrastructure strain rather than a corporate feud. Different reason, same mechanism: a model lab deciding, unilaterally and on its own timeline, who gets to keep using its models and under what terms. Cursor users who assumed “model-agnostic” meant “permanent access to every frontier model” are learning the same lesson OpenClaw users learned five months ago. The lab that trains the model controls the tap.
Anthropic, for its part, is leaning into the opening. Anthropic co-founder Tom Brown said the company will “continue to increase compute to support Claude models in Cursor” — which is either genuine support for developers caught in the crossfire, or a fairly obvious move to capture the traffic OpenAI is leaving on the table. Probably both.
Nobody who read the June coverage of this deal should be surprised. We wrote at the time that “the era of treating AI coding tools as model-neutral infrastructure is ending” and that “every credible tool is going to be aligned with a model bet within twelve months, either by acquisition or by the economics making neutrality unsustainable.” What we didn’t predict — and what’s more striking than the underlying logic — is that the alignment would arrive not from Cursor choosing to favor SpaceXAI’s models, but from a competing lab preemptively pulling out. OpenAI didn’t wait to see whether Cursor would start routing more traffic to Grok-derived models at OpenAI’s expense. It read the ownership change, read Musk’s history, and left before finding out.
That’s a meaningfully different failure mode than the one we originally flagged. The original risk was “Cursor’s new owner slowly deprioritizes rival models in favor of its own.” The realized risk was “a rival model provider unilaterally exits the moment it decides it can’t trust the new owner.” Developers planning for vendor lock-in were watching the wrong door. The one that actually closed was OpenAI’s, and it closed on OpenAI’s schedule, not Cursor’s.
It’s also worth sitting with how personal the stated reason is. OpenAI didn’t cite pricing, competitive overlap, or capacity constraints — the reasons labs usually give for winding down a partner integration. It cited Musk’s individual conduct, twice, by name, with specific incidents attached. Whatever else this is, it’s confirmation that the animosity between Sam Altman and Musk is now a live variable in enterprise contract decisions, not just a subplot in ongoing litigation.
We think OpenAI’s decision is defensible on the facts and still worth being skeptical of. A sworn admission of terms-of-service-violating distillation is a real reason to distrust a counterparty, not a pretext. Giving developers until November 12 instead of cutting access immediately is the right call, and OpenAI deserves credit for taking the slower, more disruptive-to-itself path when a faster exit was available under the contract.
That said, OpenAI isn’t a neutral party here either. Refusing to hand Astra — a model OpenAI itself has flagged for advanced cyber capability — to a company now controlled by Musk is a genuinely different kind of caution than declining to renew a coding-assistant integration over a data-licensing dispute from four years ago. Bundling both justifications into one announcement reads as OpenAI building the strongest possible case rather than the most precise one. Both things can be true: Musk’s companies do have a documented pattern of breaking agreements, and OpenAI also has every commercial incentive to avoid subsidizing a distribution channel that’s about to fund its most personally acrimonious competitor.
For developers, the practical read is simpler than the politics. If you’re running production work through Cursor’s native OpenAI models, you have about ten weeks to decide whether to move to a personal API key, lean harder into Claude or Gemini inside Cursor, or migrate the workload somewhere else entirely. That’s the same audit we recommended in June — Windsurf, Claude Code, or Google’s Antigravity are still the three credible landing spots. The difference is that the clock now has an actual date on it instead of a vague “sometime after Q3” timeline.
OpenAI says it can’t be confident SpaceX will use its models within its usage terms, citing Elon Musk’s history of contract violations at his other companies. It points to Twitter/X ending a data-licensing deal after Musk’s 2022 acquisition and to Musk’s admission under oath that xAI used distillation from competing labs’ models — including OpenAI’s — in violation of OpenAI’s terms of service.
November 12, 2026. That’s the latest date allowed under the change-of-control clause in OpenAI’s contract with Anysphere, Cursor’s parent company. OpenAI says it chose that date specifically to give developers maximum notice rather than cutting access immediately.
Not through Cursor’s native subscription billing. You can still connect a personal OpenAI API key or use OpenAI’s own IDE extensions inside Cursor, but that means paying OpenAI directly at API rates instead of through a bundled Cursor subscription.
Cursor CEO Michael Truell said OpenAI models account for roughly 5% of Cursor’s total user traffic, meaning most Cursor workloads already route to Claude, Gemini, or other models and won’t be directly affected.
No. OpenAI has said it will not provide future model releases, including its upcoming Astra model, to Cursor during the wind-down period, even though existing models remain accessible until November 12.
Yes, directly. SpaceX closed its $60 billion all-stock acquisition of Anysphere on August 14, 2026. OpenAI’s contract with Anysphere included a change-of-control clause that gave OpenAI a limited window to cancel the agreement once ownership changed, and OpenAI exercised it fifteen days after the deal closed.
Three realistic paths: switch to a personal OpenAI API key inside Cursor, shift your workflow to Claude or Gemini models that remain natively supported, or evaluate alternatives like Windsurf, Claude Code, or Google’s Antigravity if OpenAI models are central to your workflow. Ten weeks is enough time to test a migration without doing it in a panic.
Last updated: August 31, 2026. Sources: OpenAI — Our decision on Cursor following its acquisition by SpaceX · CNBC — OpenAI to end model access to Cursor after acquisition by Elon Musk’s SpaceX · The Decoder — OpenAI cuts off Cursor after SpaceX acquisition, citing Musk’s history of breaking contracts · Neowin — OpenAI terminates contract with Cursor following acquisition by Elon Musk’s SpaceX.
Related reading: SpaceX Buys Cursor for $60B: What Devs Must Know · Cursor AI Review 2026 · Windsurf vs Cursor 2026 · Cursor vs Claude Code vs Copilot 2026 · Google Antigravity vs Cursor vs Windsurf · Anthropic Cuts Third-Party Claude Tool Access · OpenAI’s AI Hacked Hugging Face — Then It Paused Astra