OpenAI says its first device remains on schedule for a 2026 unveiling. Apple’s latest answer is not a competing device but a lawsuit over what former employees may have carried into the effort.

The attention spike is abrupt, but the structure beneath it has been forming for roughly two and a half years. It began with reports that Apple design talent was leaving and has reached an argument over whether experience can move without protected knowledge moving with it.

Apple–OpenAI stories today
daily rate over the prior 90 days

The original shortcut was a team

An AI model can be distributed as software; an AI device must acquire an address. It needs materials, dimensions, controls, manufacturing decisions and a physical relationship with the person using it. The abstraction eventually becomes an object, and the fastest way to acquire the judgment required to make that object is to recruit people who have already made consequential ones.

The warning was specific by December 2023. Reports said Tang Tan, then Apple’s vice president for iPhone and Watch product design, planned to depart in February 2024. Later that month, reports said he would join Jony Ive’s LoveFrom to work on AI hardware. This was not a generic executive exit followed by an unspecified next act; the destination and domain were visible before there was a named device.

By May 2025, the network had become an organization. io was founded by Jony Ive, Scott Cannon, Evans Hankey and Tan. The sequence converted individual departures into concentrated design capability: first the people moved, then their relationships became a team, and then the team joined a hardware program with a public timetable.

The organization gave the departures coordination; the timetable gave the organization an operational target. The more credible that assembly became, the more important it became for Apple to police the boundary around its accumulated product knowledge.

The hiring strategy created its own constraint

The first phase depended on a simple distinction: companies retain their intellectual property, while people retain their general skill and experience. That distinction permits a labor market to function, but it becomes difficult to administer precisely where the talent is most valuable. Experienced engineers are recruited for judgment formed inside systems whose confidential details they cannot take with them.

Apple’s lawsuit makes that boundary concrete. According to Bloomberg’s account of the dispute, Apple says iPhone engineer Chang Liu left for OpenAI’s nascent hardware division with more than his years of experience. A separate report on Apple’s allegations says the company claims Tan, now OpenAI’s chief hardware officer and a former Apple vice president, directed Apple employees interviewing at OpenAI to share Apple secrets.

Those are allegations in litigation, not established findings of wrongdoing. The current record does not prove that Tan, Liu or OpenAI violated the law. That distinction is not a procedural footnote; it defines the present state of the contest.

The constraint nevertheless changes what counts as infrastructure. A recruiting conversation can become evidence. The provenance of a document can matter as much as its technical contents. A former employee’s contribution may need to be parsed into general expertise, remembered judgment and company-owned information—categories that appear clean in policy language and much less clean inside the work of building a product.

Hiring former Apple engineers was not inherently improper. But the strategy concentrated knowledge close enough to Apple’s product machinery that the line between portable expertise and protected information became the next contested component.

Each phase moved the scarce asset

OpenAI assembled former Apple talent because that experience could accelerate an alternative to the incumbent device model. Apple’s legal response treats institutional knowledge as the fourth scarce asset and argues that at least some knowledge surrounding that experience must remain attached to the incumbent. Mobility produced the challenger; the challenger made the limits of mobility strategically important.

A schedule can survive while the route narrows

OpenAI’s timetable is the clearest reason not to treat the lawsuit as an accomplished blockade. The company still says it is on schedule for a 2026 unveiling and plans to ship in 2027. The record does not show that litigation has stopped the program, delayed the device or prevented the hardware group from operating.

But a constraint does not need to stop a system to alter it. In engineering, a narrowed load path can continue carrying weight while becoming the location around which the rest of the structure must be organized. Here the narrowing is around recruitment, information handling and the defensibility of contributions from former Apple employees. The program continues; the route by which Apple-trained expertise enters it is now contested.

The device race no longer begins with a model, an enclosure or a launch event. It begins with a separation problem: which capabilities belong to the engineer, which belong to the former employer, and which cannot be disentangled merely because an employment badge changed.

OpenAI’s device still sits on a 2026 unveiling calendar and a 2027 shipping calendar. Apple’s answer now sits at the exit door: before the object reaches the market, it is drawing a border around what the object’s builders can carry.