OpenAI alleges ex-employee took more than 100 product-plan files to Apple

OpenAI alleges that a former employee took more than 100 files concerning its product plans before moving to Apple.² The company set out the accusation in a filing tied to a trade-secrets dispute between the two firms.² The reported count is more than 100 files, not a stated number of products, devices or projects.² OpenAI says the material included images of at least one product that had not been released, alongside supplier details, engineering information and project files.² Those descriptions identify the kinds of material at issue, but the allegation itself does not establish that Apple used any of it.¹²

According to OpenAI’s account in the filing, the former employee transferred the information to a personal account and later showed it to Apple leaders.² The people at the presentation reportedly included a vice president whose responsibilities covered operations in Asia.² OpenAI says the presentation concerning an unreleased product happened months before the filing was made.¹ That timing distinguishes the alleged presentation from the later legal document: the company says the first preceded the second by months, without specifying an exact interval in the cited account.¹² The material described in the filing ranges from product imagery to supplier and engineering information, rather than being limited to a single type of document.²

OpenAI says it is deeply concerned that confidential information could now be circulating within Apple.¹ That statement reports OpenAI’s concern, not a finding that the information spread or was used.¹² Apple rejects the allegations, says safeguards apply to people it hires, and reports that it has found no evidence supporting OpenAI’s claims.¹ The two companies therefore present opposing accounts, and the allegations remain disputed rather than resolving whether Apple used the material.¹² The filing does not establish that allegedly misappropriated information contributed to development of Home View.¹

Apple has brought a separate case that accuses former Apple employees of taking proprietary information to OpenAI’s consumer-hardware division.¹ OpenAI denies Apple’s claims that employees who left downloaded confidential files and carried knowledge of unreleased technologies into their new roles.¹ The two complaints thus describe alleged movement of protected information in opposite directions: OpenAI’s filing concerns a former employee moving to Apple, while Apple’s suit concerns former employees moving to OpenAI.¹² This comparison describes the parties’ allegations, not established transfers or proof of use.¹² The available account says the dispute concerns handling of proprietary information on both sides, while leaving the central accusations contested.²

A federal judge allowed Apple faster access to some OpenAI hiring records, including interview guides, technical presentations and documents about conversations concerning Apple technologies or components.¹ The judge also approved depositions of three former Apple employees: Tang Yew Tan, Chang Liu and Alyssa Peng.¹ Apple sought records to investigate whether its proprietary technology featured in OpenAI’s recruitment process.¹ The judge did not grant Apple’s wider requests for OpenAI hardware research and certain engineering information.¹ These steps concern records and testimony relevant to the dispute; they do not themselves resolve the allegations about what information was taken or used.¹²

The dispute over information moving between the companies sits alongside a separate change in OpenAI’s computing leadership.³ Infrastructure chief Sachin Katti said he would leave that position after about a year and remain an adviser after November 15.³ OpenAI’s spokesperson said Uday Ruddaraju, its chief technology officer for compute, would replace Katti and lead the company’s overall compute organization.³ The reported duration, about a year, describes Katti’s time in the role, while the succession account identifies Ruddaraju’s incoming responsibilities.³ The available facts do not connect this leadership transition to the trade-secrets allegations, so the two developments remain distinct in this account.²³

OpenAI’s mathematics work provides another, separate part of the reporting about the company.⁴⁵ One source says OpenAI placed 722 AI-generated mathematics manuscripts on GitHub on October 6, 2026, and attributes them to an unreleased internal frontier model.⁴ That account counts 372 result families and 377 new proofs among the manuscripts, and says about 42% of their main conclusions had passed Lean formal verification.⁴ A second report also gives a total of 722 manuscripts, but places the upload on October 7 and says some Lean proofs were included.⁴⁵ The two accounts agree on the manuscript total but differ on the stated release date, so the dates should not be collapsed into one.⁴⁵

The QbitAI report focuses on two manuscripts about the Kakeya problem, which together ran to 272 pages and named OpenAI as their only author.⁵ It says those two papers had not undergone Lean formalization or peer review.⁵ That report’s statement about the Kakeya papers is narrower than the other source’s approximate verification figure for the main conclusions across the manuscript collection.⁴⁵ The measures are not interchangeable: one describes an approximate share of conclusions passing Lean checks, while the other addresses formalization and peer review of two specific papers.⁴⁵ The available accounts therefore describe limited verification in different ways, rather than providing one uniform assessment of all 722 manuscripts.⁴⁵

Taken together, the reporting covers a contested legal dispute, court-approved access to selected hiring materials, a computing leadership handoff and a large mathematics release.¹²³⁴⁵ The legal accounts identify allegations by both companies and describe procedural steps, but they do not establish that either side used the information at issue.¹² The research reports provide counts and verification qualifications, including different dates for the public upload and distinct methods of describing review.⁴⁵ Keeping those categories separate matters to reading the record: an allegation, a judicial decision about access, a company statement and a reported research count each describe different things.¹²⁴