Musk Dropped Apple From the Suit. A Judge Wants to Read the Deal by Noon Today, Because OpenAI Is Still a Defendant.

On 14 September 2026, X Corp. and SpaceXAI LLC moved to dismiss their antitrust claims against Apple with prejudice. They told the court they had resolved their claims. Apple did not oppose. The filing offered no explanation and did not say whether a settlement existed. Claims against the OpenAI defendants continue.

That is ordinarily the end of a news cycle rather than the start of one. A plaintiff releasing one defendant, unopposed by the defendant being released, is nobody else's business. This one did not go quietly.

On Tuesday 15 September, US District Judge Mark Pittman, in the Northern District of Texas at Fort Worth, ordered the companies to deliver for in-camera review any agreement or combination of agreements with Apple related to resolving the claims. The deadline is noon today, Thursday 17 September. Apple may respond by the same deadline but is not compelled to. The order followed an emergency motion from OpenAI, which told the court it did not participate in the agreement and argued its terms could affect OpenAI's defence.

One procedural detail is genuinely unclear from the public record and we flag it rather than resolve it: the docket shows a 14 September entry captioned as an order of dismissal or administrative closure, and at least one report described Pittman as having granted the voluntary dismissal on Monday. Whether the dismissal was entered before the in-camera order, or remains pending, is not something secondary coverage settles. The analysis below holds either way, and the question of what the agreement contains arguably matters more, not less, if the dismissal has already been entered.

Why the defendant who stayed gets a say

The instinct is that OpenAI has no business reading a contract between two other parties. The reason that instinct is wrong is specific to antitrust, and it is the mechanism worth understanding.

Start with what federal antitrust law does not provide. In Texas Industries, Inc. v. Radcliff Materials, Inc., 451 U.S. 630 (1981), the Supreme Court held that there is no right of contribution among antitrust defendants. A defendant who pays a judgment cannot turn around and sue a co-defendant for its share. Combined with joint and several liability, that means a single remaining defendant can be made to pay the entire measure of a plaintiff's antitrust damages, which under the Clayton Act are trebled.

What a non-settling defendant gets instead is a settlement credit. The plaintiff's recovery against the remaining defendant is reduced by the amount the settling defendant paid, typically dollar for dollar on the pro tanto approach, before trebling is applied. That credit is the only protection available, and its size is set entirely by a transaction the remaining defendant had no part in negotiating.

Which is why the amount matters to OpenAI, and why the characterisation matters as much as the amount. A payment structured as consideration for a release of these claims reduces OpenAI's exposure. The same money characterised as something else, a commercial arrangement between Apple and the plaintiffs, say, may not. The two look similar in a press release and are worth very different sums to the defendant left in the case.

The part that is not about money

Settlement agreements in multi-defendant litigation routinely carry non-monetary terms, and these can matter more than the number. Cooperation provisions commit the settling party to produce documents, make witnesses available, or assist the plaintiff's presentation of the remaining case. A co-defendant that was, until last week, aligned against the plaintiff on shared facts can emerge from a settlement as something closer to a resource for the other side.

There is also the inverse: terms that restrict what the settling party will say or produce, which can make evidence that was reachable through a co-defendant's ordinary discovery obligations harder for the remaining defendant to obtain. Either direction reshapes the case OpenAI has to try.

Put together, the picture is a defendant whose maximum financial exposure is set by a credit it cannot negotiate, and whose evidentiary position can be altered by commitments it cannot see. That is a coherent basis for an emergency motion, whatever one thinks of the underlying suit.

What in-camera review is for

Once a defendant has answered, a plaintiff cannot simply walk away unilaterally. Federal Rule of Civil Procedure 41(a)(2) provides that dismissal at that stage is by court order, on terms the court considers proper. That phrase is the whole hook. It gives a judge discretion to look at what a dismissal does to parties other than the two agreeing to it.

In camera means the judge reads the document in chambers. It is not filed publicly, and the co-defendant does not automatically see it. The point of the mechanism is to let a court answer a narrow question, whether releasing this defendant prejudices the one who remains, without publishing terms that are commercially sensitive and that the parties have every ordinary right to keep private.

The realistic outcomes are a short list. Pittman can grant the dismissal as filed. He can grant it with conditions, for instance by preserving OpenAI's ability to claim a credit or to seek discovery of specified terms later. He can order limited disclosure to OpenAI under a protective order, which is the middle path courts often take. Or the production can reveal there is no written agreement of substance at all, in which case the matter closes without incident.

It is worth saying plainly: ordering production is not a finding that anything improper occurred. Judges request documents in camera precisely when they do not know, and the order is better read as a court declining to rubber-stamp than as a court suspecting misconduct.

The case that this is routine

The plaintiffs' and Apple's position deserves stating at full strength, because it is not weak. Defendants settle out of multi-defendant cases constantly, and the remaining defendants almost never get to read the agreement. The settling defendant here does not oppose its own release. The credit that protects OpenAI is computed at judgment, if there ever is one, not at the moment of dismissal, so OpenAI's interest is on this view not yet ripe and the emergency framing is overstated. A co-defendant has no general standing to audit the terms of a contract it was not party to, and treating every settlement as reviewable would make multi-party litigation substantially harder to resolve.

Courts frequently accept exactly that reasoning. The fact that this judge did not, at least far enough to look, is the newsworthy part.

What today's deadline does and does not settle

The most likely outcome of noon today is that the public learns nothing. In-camera means in-camera. If Pittman reads the agreement and is satisfied, the docket may record little more than that the order was complied with and the dismissal stands. Whether money changed hands, how much, and what either side committed to do could remain permanently unknown outside chambers and the parties.

Every inference about what Apple gave or got is therefore speculation, and readers should discount confident accounts accordingly, including accounts framing this as a Musk-Apple rapprochement. There is no public evidence about the commercial relationship between these parties beyond the bare fact that claims were resolved.

What today does settle is narrower and still useful. It establishes whether a written agreement exists. It puts the terms in front of a judge who has already shown he will not treat this dismissal as a formality. And it creates a record that OpenAI can point to later when the question of settlement credit becomes live.

The broader shape is worth noting. X Corp. and X.AI LLC filed this suit on 25 August 2025, alleging that Apple's integration of ChatGPT into Siri and Apple Intelligence advantaged OpenAI over rival chatbots including Grok. In August 2026 the court granted an unopposed request to change the second plaintiff's caption name to SpaceXAI LLC. In November 2025 Pittman denied both defendants' motions to dismiss, letting the case proceed past the pleadings.

After Apple's exit, what began as a two-defendant distribution case is a single-defendant case against OpenAI, before a judge who has already declined to throw it out once, with a settlement credit of unknown size sitting between OpenAI and whatever damages figure a plaintiff eventually argues for. The case did not shrink. It concentrated.