Apple's July 9 filing accuses a former engineer of using a bug in its build system to exfiltrate Core ML compiler source, the code that turns trained models into instructions Apple's own chips can run, and hand it to OpenAI. GPT-5.6, shipped the same week, is the model this suit says benefited. The Verge's read of the complaint calls it Apple's wildest claim yet against a rival, but the wildness is beside the point. What matters is the fact pattern: a named engineer, a named exploit path, a named recipient. Apple isn't arguing OpenAI copied a model's outputs, the usual shape of an AI copyright fight. It's arguing OpenAI received the tooling that compiles models onto Apple's own silicon, the layer that decides whether a model runs fast and cheap on an iPhone or drains the battery in twenty minutes. That's a supply chain claim, not a plagiarism claim, and it opens discovery into whether GPT-5.6's on-device efficiency owes anything to code Apple never licensed.

The GPT-5.6 launch also brought unrelated file-deletion reports serious enough that TechCrunch tracked repeated user warnings, and OpenAI shipped it into a company still missing an AGI chief, a safety chief, and now a president, all vacated without successors this quarter. A model behaving unpredictably with local file access is a reliability problem any lab can hit. A model like that shipping through a management structure with three open executive seats is a governance problem, because there's no named owner left to decide whether to pause the rollout while the Apple discovery process runs. DeepMind's Demis Hassabis used the same week to call for an independent standards body to audit frontier labs, a proposal that reads less like foresight and more like a description of what OpenAI's own leadership bench can't currently do internally.