Apple Sues OpenAI Over Trade Secret Theft

OpenAI’s hardware push collides with Apple in federal court, its head of safety exits, and Greg Brockman takes control of product ahead of the lab’s expected IPO.

July 11, 2026 | Reading time: 11 minutes | Issue #207

Lead

Apple sued OpenAI in federal court in Northern California on Friday, alleging that the AI lab systematically stole confidential information from the iPhone maker to build its own consumer hardware. The complaint, filed in the U.S. District Court for the Northern District of California, names OpenAI, its hardware subsidiary io Products, OpenAI Chief Hardware Officer Tang Tan, and former Apple senior systems electrical engineer Chang Liu as defendants. Apple says the scheme ran from individual technical staff up to the chief hardware officer, and that it involved job candidates bringing Apple parts to interviews, departing employees bypassing exit security, and unauthorized downloads of confidential files.

The central figure is Tan, who spent 24 years at Apple before joining OpenAI and co-founding io Products, the device vehicle OpenAI acquired for roughly $6.4 billion in 2025. Apple alleges Tan used internal Apple codenames during recruiting and asked candidates to bring "actual parts" — batteries, logic boards, and SIPs — to "show and tell" sessions. Chang Liu, the second named defendant, is accused of retaining a work-issued Apple laptop after leaving and exploiting a bug to access Apple cloud storage while employed by OpenAI, downloading dozens of files labeled confidential. Apple says it sent OpenAI a letter in February and received no response.

The lawsuit arrives as OpenAI prepares a hardware device that analysts, including Ming-Chi Kuo, have speculated could be an AI-first smartphone. OpenAI CEO Sam Altman has said the first prototypes are finished. Apple and OpenAI still have a partnership integrating ChatGPT into Apple Intelligence, which makes the timing more awkward. OpenAI responded on Friday: "We have no interest in other companies' trade secrets. We remain focused on building innovative technology that empowers people everywhere." Apple is seeking damages, injunctions, and an order forcing OpenAI to stop using its trade secrets. The case will proceed through discovery, where Apple may learn how deeply its information shaped OpenAI’s hardware program.

OpenAI’s Safety Chief Departs

OpenAI’s head of safety systems, Johannes Heidecke, told staff this week that he is leaving the company, WIRED reported. His departure follows a reorganization in which OpenAI’s safety teams now report to vice president of research and alignment Mia Glaese, who becomes the company’s VP of research and safety. Saachi Jain, who previously led safety teams, will serve as interim head of safety systems.

Heidecke joined OpenAI in 2021 and took over as head of safety systems in 2024 after Lilian Weng left to co-found Thinking Machines Lab. Chief Research Officer Mark Chen wrote in a staff memo that safety coordination challenges have grown because models are being trained faster and release cycles have shortened. The same week, OpenAI chief futurist Joshua Achiam also said he was leaving after nearly nine years, and CEO of AGI deployment Fidji Simo stepped down from her role after an extended medical leave.

The departures come as OpenAI is rolling out GPT-5.6, which the company has acknowledged showed "concerning forms of misaligned behavior" in internal evaluations. The pattern is a governance problem as much as a technical one: the people responsible for flagging misalignment are leaving just as the models are becoming more capable and the release cadence is accelerating.

Greg Brockman Consolidates Control

OpenAI co-founder Greg Brockman is now officially responsible for the company’s most important product and business projects, CNBC reported. He took over product responsibilities during Fidji Simo’s medical leave and will continue to lead ChatGPT, go-to-market, enterprise, and compute initiatives. Simo, the former Meta and Instacart executive, is transitioning to a part-time advisor role.

Reporting directly to CEO Sam Altman, Brockman is under pressure to generate revenue and justify OpenAI’s roughly $852 billion valuation as the company prepares for a widely expected IPO. OpenAI confidentially filed its prospectus in June. ChatGPT’s market share fell below 50% for the first time in March, according to Sensor Tower, and OpenAI has been pushing its Codex coding assistant to broaden its user base. The company is not planning to hire a replacement for Simo, according to a person familiar with the plans, meaning Brockman’s authority is not temporary.

Meta Suspends Muse Image After Backlash

Meta discontinued a feature that let users generate AI images by tagging public Instagram accounts, following criticism from CAA, SAG-AFTRA, and privacy groups. The feature, part of the new Muse Image model, did not notify users when their likeness was referenced and required account holders to opt out. CAA, whose clients include Tom Hanks and Meryl Streep, called it an infringement on name, image, and likeness rights. SAG-AFTRA told members to opt out. Public Citizen called it an "egregious invasion."

Meta’s reversal was swift: the feature was introduced on Tuesday and removed by Friday. A spokesperson said the tool "missed the mark." The episode shows how fast consent disputes can escalate for consumer AI products, especially when the default is opt-out rather than opt-in. It also underscores the risk that platforms with billions of users can turn a single feature into a de facto likeness-scraping engine before regulators or the public can react.

Open-Source Pulse

Tencent’s Hunyuan team released Hy3, a 295-billion-parameter Mixture-of-Experts model with 21 billion active parameters, under Apache 2.0, continuing the week’s open-weight momentum. Earlier Chinese open-weight releases often excluded the EU, UK, and South Korea, which complicated enterprise adoption. Hy3 removes that obstacle and will be free on OpenRouter for two weeks.

Meanwhile, U.S. lawmakers are probing the growing use of Chinese AI models by American companies. The House Committee on Homeland Security and the House Select Committee on China sent letters to Cursor and Airbnb over their exposure to Chinese-developed models. Cursor, which is being acquired by Elon Musk’s SpaceX for $60 billion, built its Composer 2 model using Moonshot AI’s Kimi. Brookings fellow Kyle Chan noted that banning open-source Chinese models is effectively impossible because their weights are freely available online, raising First Amendment questions. The committees are examining whether the U.S. has a sufficient open-weight strategy to keep American companies from defaulting to cheap, capable Chinese alternatives.

Eastern Front

Chinese AI models are gaining ground among U.S. companies precisely because they have narrowed the performance gap while undercutting American rivals on price. The House probe cites concerns that Chinese models carry embedded censorship, uncertain security, and capabilities distilled from U.S. laboratories with safety guardrails stripped out. A State Department spokesperson told CNBC that Chinese AI models are "designed to advance Beijing’s narratives, censor dissent, and reflect CCP ideology and values." Beijing, in turn, is reportedly considering limiting overseas access to China’s leading AI models, a move that would tighten the global supply of open-weight models and force U.S. companies to make more explicit procurement choices.

Compute Watch

Meta announced it is building its first large data center in Canada, a 1 gigawatt facility in Alberta that will cost approximately $9 billion and take two to three years to construct. The site was chosen for available energy, a friendly regulatory environment, and existing industrial zoning. The project will support more than 3,000 construction workers at peak and is part of Meta’s plan to spend up to $145 billion on capital expenditures this year.

Meta is also planning a cloud computing business that could sell excess capacity or offer access to AI models hosted in its infrastructure. Investors have been skeptical of Meta’s AI spending because the company has fallen behind OpenAI, Anthropic, and Google on model quality and has not shown a clear revenue path beyond advertising. The Canadian facility is a bet that scale in infrastructure can compensate for lagging software differentiation.

Europe

QuantumDiamonds, a three-year-old Munich spin-out of the Technical University of Munich, raised €91 million to scale a diamond-based chip inspection tool. The funding includes €15 million in equity led by World Fund and €76 million in state aid approved under the European Chips Act, making QuantumDiamonds the first startup to win manufacturing funding under that program. The company’s QDm.1 uses defects in lab-grown diamonds to sense magnetic fields and image currents in 3D at the nanoscale, allowing it to locate buried defects without destroying the chip. It says it already works with nine of the world’s ten largest chipmakers.

The funding fits Brussels’ broader push to reduce reliance on U.S. and Asian semiconductor supply chains. World Fund managing partner Daria Saharova cast the company as a potential national champion, calling it a candidate to become "Europe’s next ASML." The comparison is premature — QuantumDiamonds has one product and must convert trials into paid production contracts — but it captures the strategic mood in Europe around sovereign chip capability.

India Lens

India’s Ministry of Electronics and Information Technology issued a notice to Meta demanding that Instagram disable paid advertisements promoting child sexual abuse material and submit a detailed explanation within seven days. The directive followed a BBC investigation that found such ads running in India. Meta said it has taken action against the violating accounts and ads and is cooperating with law enforcement. The incident is the latest in a series of regulatory confrontations in Meta’s largest market by users, where its content moderation systems are being tested by scale, local language content, and the proliferation of AI-generated material.

From the Lab

OpenAI released GPT-5.6 on Thursday in three tiers — Sol, Terra, and Luna — after the Trump administration lifted earlier rollout restrictions. Sol is the flagship, Luna is optimized for speed, and Terra balances the two. CEO Sam Altman told CNBC that Sol is 54% more token-efficient on agentic coding tasks than Anthropic’s latest model. The company also introduced ChatGPT Work, an agent that can gather context across connected apps to create documents, spreadsheets, and presentations.

On arXiv, researchers introduced UniClawBench, a benchmark for proactive agents operating in dynamic, real-world settings with 400 bilingual tasks evaluated in live Docker containers. A separate paper, OpenCoF, explores Chain-of-Frame reasoning through video generation, and IdeaGene-Bench tests whether models can reason about scientific idea inheritance. The research suggests the field is shifting from static benchmark scores toward evaluations that better isolate why agents fail in production.

The View

Three threads run through the week. Apple’s lawsuit frames the hardware shift in AI as a zero-sum fight over talent and trade secrets, not just model capability. OpenAI’s executive turnover shows the difficulty of maintaining safety oversight while accelerating releases and preparing for public markets. Meta’s quick retreat on Muse Image demonstrates that consumer AI features can collapse under consent pressure faster than they can be shipped.

The common pressure is time. OpenAI needs to ship hardware and justify an $852 billion valuation before competitors close the gap. Meta needs to show AI revenue before investor patience thins. Apple needs to defend its industrial secrets as its AI partners become rivals. The result is a market where legal, regulatory, and reputational risks are piling up alongside technical ones. Companies that can slow down enough to verify safety, secure data, and obtain consent may find themselves moving faster than those that rush past those steps and stumble.

The Miss

OpenAI’s "Separating Signal from Noise" audit of SWE-Bench Pro argued that roughly 30% of tasks in the widely used coding benchmark are broken — overly strict tests, underspecified prompts, low-coverage tests, and misleading instructions. The paper received less attention than model launches, but it matters because flawed benchmarks shape hiring, research priorities, and safety claims. If the field is optimizing for broken tests, the leaderboard leaders may not be the most useful systems.

Pull Quotes

"At every level, from members of its Technical Staff to its Chief Hardware Officer, and in coordination with business partners, OpenAI has been stealing Apple’s trade secrets and confidential information." — Apple legal filing, July 10, 2026.

"We have no interest in other companies’ trade secrets. We remain focused on building innovative technology that empowers people everywhere." — OpenAI statement, July 10, 2026.

"The demands on safety continue to increase — we are training models at a much faster cadence, and release cycles have come down greatly." — Mark Chen, OpenAI chief research officer, WIRED, July 10, 2026.

"No one’s name, image, likeness, voice or creative work should be used by any third party, including AI models, without clear, documented consent." — Creative Artists Agency, Variety, July 10, 2026.

Reads & Links

Apple sues OpenAI alleging trade secret theft — CNBC, July 10, 2026. https://www.cnbc.com/2026/07/10/apple-openai-lawsuit-trade-secrets.html

Apple sues OpenAI over trade secrets — Axios, July 10, 2026. https://www.axios.com/2026/07/10/apple-sues-openai-trade-secret-theft

OpenAI’s Head of Safety Is Leaving the Company — WIRED, July 10, 2026. https://www.wired.com/story/openai-head-of-safety-leaving/

OpenAI power consolidates under co-founder Greg Brockman ahead of prospective IPO — CNBC, July 10, 2026. https://www.cnbc.com/2026/07/10/openai-power-consolidates-under-co-founder-greg-brockman-ahead-of-ipo.html

Meta Suspends Instagram AI Image Feature After Days of Backlash — Variety, July 10, 2026. https://variety.com/2026/biz/news/meta-suspends-ai-image-instagram-feature-backlash-1236806989/

Lawmakers probe growing use of Chinese AI models in U.S. companies — CNBC, July 8, 2026. https://www.cnbc.com/2026/07/08/chinese-ai-models-probe-us-lawmakers.html

Meta is building its first big Canadian data center as AI expansion crosses the border — CNBC, July 8, 2026. https://www.cnbc.com/2026/07/08/meta-is-building-its-first-big-data-center-in-canada-amid-ai-push.html

A €91M bet on the diamond microscope that could be Europe’s next ASML — The Next Web, July 10, 2026. https://thenextweb.com/news/quantumdiamonds-91m-eu-chips-act-inspection

France’s Skello secures €200 million to grow its AI tools for frontline workforce management — EU-Startups, July 6, 2026. https://www.eu-startups.com/2026/07/frances-skello-secures-e200-million-to-grow-its-ai-tools-for-frontline-workforce-management/

OpenAI releases GPT-5.6 and ChatGPT Work tool — Axios, July 9, 2026. https://www.axios.com/2026/07/09/ai-openai-gpt-release

UniClawBench: A Universal Benchmark for Proactive Agents on Real-World Tasks — arXiv, July 9, 2026. https://arxiv.org/abs/2607.08768v1

Out

The next frontier is not a model; it is the courtroom, the boardroom, and the opt-out setting.