OpenAI's Stormy Legal Week. Safety Keeps Losing Its Seat.

OpenAI is facing a series of legal challenges with potential capital-markets implications. Over the past week, they've been sued by Apple for alleged trade secret theft, and a federal judge was asked to sanction them for their conduct in the ongoing New York Times copyright case. These pressures coincide as the company moves toward an IPO it confidentially filed for last month.

On July 10, Apple sued OpenAI in the Northern District of California for trade-secret theft and breach of contract. The complaint names Tang Tan, OpenAI's Chief Hardware Officer — who spent 24 years working at Apple before leaving to build OpenAI's hardware effort — and Chang Liu, a former Apple engineer. Apple also named io Products, the hardware startup OpenAI acquired for $6.5 billion in 2025, as a defendant. The suit accuses them of systematically taking confidential hardware information from Apple and using it to develop OpenAI's unreleased device.

Apple is seeking a preliminary injunction that would require OpenAI to preserve evidence, stop using Apple technology, and return the trade secrets. If granted, that order lands directly on the hardware business the acquisition was meant to create. It's worth remembering that Apple and OpenAI announced a partnership in 2024 to put ChatGPT inside iOS. Two years later, and one partner is suing the other.

The suit itself is not new — The NYT originally filed back in December of 2023, alleging that OpenAI (and co-defendant Microsoft) trained its models on NYT articles without permission or a license, and that the resulting products could reproduce Times journalism with such high accuracy that they could act as a substitute for the original reporting, and siphon revenue away from the Times.

On July 9, the Times and a group of 16 other news organizations asked the court to sanction OpenAI, alleging that it spent two years falsely claiming it couldn't search its own training data and output logs. The publishers point to deposition testimony indicating that OpenAI had in fact been conducting searches of those systems, which would conflict with its earlier representations to the court. The motion also alleges that OpenAI deleted or made unsearchable billions of ChatGPT conversations, continuing to do so after the court had ordered the relevant output data preserved.

Two adversarial legal fronts escalating at once, one alleging trade secret theft and the other pairing a copyright claim with an accusation of bad-faith discovery. These are exactly the kinds of risks that surface in an S-1 risk-factors section, and OpenAI has already filed confidentially, with a public listing still ahead.

OpenAI has opened the door to a public listing without committing to a timetable. Even so, litigation that intensifies before a potential IPO can shape disclosure, investor perception and the valuation narrative.

OpenAI's head of safety systems, Johannes Heidecke, will leave the company by July 24. He is the latest in a string of senior departures from OpenAI's safety and alignment functions. In February, the head of Anthropic's safeguards team resigned, leaving a message that "the world is in peril". xAI has meanwhile experienced a broad co-founder exodus. On their own, these aren't remarkable events, but read together, they describe a trend that we're seeing across the industry: the safety function is steadily losing its independent seat as the frontier continues to advance.

At OpenAI it's becoming something of a pattern. The Superalignment team, co-led by Ilya Sutskever and Jan Leike dissolved within days of both leaders announcing their resignations in May 2024. AGI-readiness lead Miles Brundage — who led a team meant to prepare the company and society for increasingly capable AI — left that same year in October. The Mission Alignment unit, formed in September 2024 to take over from the Superalignment team, was disbanded in February 2026 and its members redistributed into research and product. Heidecke's departure now folds safety systems into research as well: safety teams will report to Mia Glaese, elevated to VP of Research and Safety, with Saachi Jain as interim head. Chief Research Officer Mark Chen framed the move as integrating safety more directly into model development, but a more skeptical reading is that safety oversight is steadily losing its organizational independence.

Anthropic is perhaps an even more striking case — a lab that markets itself on safety. On February 9, Mrinank Sharma — who joined Anthropic in 2023 and led its Safeguards Research Team from launch — resigned in a letter posted publicly to X. He wrote that he had repeatedly seen how hard it is to let values govern actions, both in himself and inside an organization where, in his words, "we constantly face pressures to set aside what matters most". In his public resignation letter he warned that, "the world is in peril", pointing to a collection of interconnected threats that included — but were not limited to — AI.

On July 7 the Future of Life Institute (FLI), an AI-risk advocacy organization, published its Summer 2026 AI Safety Index, using six domains to grade nine developers. Under the institute's methodology, no company scored above a C+. xAI, DeepSeek, and Mistral failed outright. One of the index's central findings was that many of these developers, including Anthropic, OpenAI, DeepMind, and Meta have either weakened or dropped entirely earlier pledges to pause development if their systems approached dangerous thresholds. The research panel called this, "moving the goalposts".

If you are running diligence on an AI vendor or an AI target, the safety function is becoming harder to read. FLI's panel findings suggest that the major AI labs are weakening or dropping their commitments to pause development at specific risk thresholds. Those pledges were voluntary. Whatever weight they carried six months ago, they carry less now.

The org charts seem to be mirroring the story we're seeing in the amended policy commitments: as the models get more capable, the release cycles get shorter, and the competition between labs grows more intense, the policies governing them weaken while the people responsible for keeping their development safe grow scarcer.