OpenAI vs. Apple Heats Up as Astra Cracks Hard Math

Two distinct stories dominated OpenAI coverage on August 4: a rapidly escalating legal fight with Apple, and a striking mathematical research claim from the company’s new Astra model family.

Key points

• Apple has filed a motion for a preliminary injunction against OpenAI, seeking to stop former Apple employees now at OpenAI from using confidential information.
• Apple also filed a motion for expedited discovery, aiming to accelerate access to relevant documents.
• OpenAI pushed back publicly, calling the lawsuit meritless and claiming it has documentary evidence to support its position.
• OpenAI’s Astra model produced machine-checkable proofs for ten longstanding mathematical problems at a reported compute cost of $2,000, with the work published August 1.
• The math results span group theory, high-dimensional geometry, coding theory, quantum complexity, lattice cryptography, and combinatorics, several of them open for decades.

What is Apple actually asking for?

Apple’s July complaint alleged that former employees who moved to OpenAI brought confidential information with them, misappropriating trade secrets tied to the company’s future hardware development. The preliminary injunction motion would compel both the named ex-employees and OpenAI itself to stop using any such information while the case proceeds. Apple also wants expedited discovery, which, if granted, would force OpenAI to produce documents faster than in standard litigation timelines.

The framing from Apple’s side is that competitive harm is ongoing and urgent. Preliminary injunctions require a plaintiff to show a likelihood of success on the merits and irreparable harm, a high bar. Apple’s move to seek one this early signals confidence in its evidentiary position, or at minimum a desire to apply maximum pressure quickly.

How is OpenAI responding?

OpenAI’s counter was pointed. The company denied any wrongdoing and stated it has no interest in Apple’s trade secrets. Crucially, it indicated it has messages or other documentation to support that position, framing Apple’s complaint as baseless. The phrase “has receipts” in coverage suggests OpenAI may be planning to introduce contemporaneous communications as evidence.

This is early-stage litigation posturing, and neither side’s public statements should be read as indicative of the legal outcome. What matters for investors watching OpenAI’s IPO trajectory is that the company now faces a high-profile adversarial legal proceeding with one of the world’s largest technology companies, layered on top of ongoing regulatory scrutiny in other areas. Discovery alone can be expensive and disruptive.

What did Astra actually do in mathematics?

OpenAI published a 249-page manuscript on August 1 claiming its Astra model generated proofs for ten problems that had resisted human progress for at least a decade each. The list includes a construction proving the existence of non-sofic groups, a disproof of Connes’s rigidity conjecture in von Neumann algebra theory, an improvement to the upper bound on sphere-packing density in high dimensions (a bound that had stood since 1978), and three problems from the open-question catalogue left by mathematician Paul Erdos.

The reported compute cost was $2,000. More notable than the price tag is the verification method. OpenAI attached a machine-checkable certificate to every result, something coverage describes as unprecedented at this scale for an AI-generated math claim. That matters because machine-verified proofs are, in principle, auditable by independent mathematicians using formal verification tools, reducing the scope for the kind of undetected errors that have plagued prior AI math announcements.

This follows a May result in which OpenAI used a similar reasoning model to disprove a geometry conjecture first posed by Erdos in 1946. OpenAI is using the Astra name for its next major model family, tying the brand to this research output.

Some mathematicians have already raised general concerns about AI-generated research flooding the field, a caution worth keeping in mind as independent review of the August 1 manuscript proceeds. The claims are significant enough that scrutiny will be intense.

Why this matters for OpenAI’s broader position

The two stories pull in opposite directions. The Astra math results, if they hold up to peer review, strengthen OpenAI’s case that frontier AI delivers genuinely novel scientific value, a useful narrative for both enterprise customers and future public-market investors. The Apple litigation, by contrast, adds legal and reputational risk at a moment when OpenAI is already navigating questions about its corporate structure and governance.

Neither development changes OpenAI’s near-term business fundamentals in an obvious way. But both reinforce that the company is operating simultaneously at the cutting edge of capability and the center of an expanding set of legal disputes.