Apple Accuses OpenAI's Leadership of Running the Theft; Proctored Finals Cut Brown Grades in Half

01Apple Says OpenAI's Own Leadership Ran the Theft, Not a Rogue Engineer

Apple sued OpenAI on Thursday, and the complaint aims higher than a single departing engineer. Apple alleges the theft of its hardware trade secrets was directed by OpenAI's senior leadership, including a long-time former Apple employee now working at the startup. The framing matters. Companies typically sue over one person who walked out with files; Apple is describing a coordinated effort authorized from the top.

In its filing, Apple says it uncovered "a pattern of theft of Apple's trade secrets by OpenAI employees who were formerly at Apple," according to the complaint. The word "pattern" does the work here. Apple is not pointing to a lone document or a single hire. It alleges repeated conduct by multiple ex-Apple staff, tied together by direction from people running the company.

The lawsuit does not stop at OpenAI. It also names IO Products, the hardware startup founded by Jony Ive, the former Apple design chief. That places two of Apple's former insiders on the other side of the case: the design leader who shaped its most valuable products, and an unnamed long-time employee Apple says helped steer the alleged theft. Both now sit inside OpenAI's push into physical devices.

The two outlets covering the suit frame it with different emphasis. TechCrunch leads with the direction from senior leadership, the claim that turns a personnel dispute into an institutional one. The Verge foregrounds the mechanics: engineers who took Apple secrets to advance OpenAI's hardware plans, and the inclusion of Ive's IO Products as a co-defendant. Read together, the accounts describe the same core allegation from two angles, one about who ordered it and one about who carried it out.

The two companies were partners as recently as their consumer-facing integration work, which makes the venue notable. Apple guards hardware secrets more tightly than almost any company its size, and OpenAI is moving fast to build physical products it has never shipped. Those two positions were always going to grind against each other once the same people worked on both sides.

OpenAI has not filed its response, and none of Apple's allegations have been tested in court. Apple has asked the court to hold OpenAI and IO Products liable for the alleged misappropriation.

02A Solar Company Will Pay You to Keep an AI Data Center in Your House

Sunrun, a solar and home battery installer, is starting a pilot that inverts how data centers get built. Instead of raising billions to pour concrete and pull grid connections, the company says it will pay its own customers to host "compute nodes" inside their homes. Sunrun calls it "distributed AI compute," and it plans to place the units in residences already wired with its solar panels and storage.

The logic is physical, not ideological. Conventional data centers wait years for utility interconnections and megawatt-scale power hookups. Sunrun's fleet of home solar-and-battery systems already sits behind the meter, generating and storing electricity on site. Putting compute where the power already exists skips the queue that has become the binding constraint on AI expansion.

That constraint is visible in the emissions of the companies with the deepest pockets. Microsoft's 2026 sustainability report, first reported by GeekWire, states that its carbon emissions rose 25 percent in 2025, reaching 34 million metric tons "without select interventions." The company attributes the increase primarily to the expansion of its infrastructure. It is the second straight period in which Microsoft has reported moving away from its own 2030 carbon-negative pledge.

Microsoft's number measures the cost of building at scale inside the grid. Sunrun's pilot measures what happens when builders stop waiting for the grid at all. Both point at the same ceiling: the standard playbook of large campuses drawing utility power is running past what interconnection timelines and clean-energy commitments can absorb.

Sunrun has not disclosed how many homes the pilot will reach, what it will pay hosts, or which customers run the workloads. The company frames the nodes as a way to monetize hardware its customers already own. Homeowners would be taking commercial computing equipment, and its heat, noise, and liability, into their living space in exchange for a payment Sunrun has not specified.

03A Brown professor moved the final into a proctored room. Grades dropped by half.

When a professor at Brown University suspected students were leaning on AI to finish his coursework, he changed one variable. He swapped the take-home final for an in-person, closed-book exam. Scores fell by roughly 50 percent.

The number worked as a controlled measurement. Under the old format, the grades reflected whatever students plus their tools could produce. Sit them in a room with no assistant, and half the performance disappeared. What got outsourced was not busywork. It was the part that showed up on the transcript as competence.

The professor did not soften the result. AI cheating, he said, leads to "a failed society." His line for the students was blunter: "we cannot choose to become idiots." He was describing a capability that erodes when the machine does the reps instead of the person.

A developer writing under the title "I think I have LLM burnout" describes the same trade from the inside. He uses LLMs for hours every day, by his own account an average rate for a working engineer. He runs one task at a time, talking it through with Claude Code at the office and Codex at home. Sometimes he lets the assistant write the code. He says he still reads the output, understands it, and revises it.

His account of the job itself is the tell. The work used to be "designing and writing code." Now it is designing code, describing the design to an LLM, reviewing what the LLM produces, and only then writing. His current project is a framework for large-scale, unsupervised code generation, which leaves him sifting through the output of an autonomous agent running Qwen. Either way, he writes, he is reading LLM content.

For casual questions he asks ChatGPT or reads Gemini's summary, falling back to a browser only when the answer is wrong. The exposure runs constant. He does not go a day without reading AI-generated text, and the burnout in his title is what accumulates when the thinking gets handed off and the human stays on the hook for the review.