The Great AI Heist
Apple, OpenAI, and Anthropic are all accusing each other of stealing the future. so Whats exactly is happening!
It looks like everyone is accusing everyone else these days. from stolen iPhone secrets to scraped chatbot answers, the biggest names in AI are suddenly treating “open” as a crime scene.
If your timeline also feels like “wow, did everyone in AI decide to sue everyone else this week?”, you’re seeing what I am seeing.
The latest in this series is, Apple suing OpenAI for allegedly stealing hardware secrets. Earlier we had Anthropic accusing Chinese labs of copying their model at massive scale. and If you remember, OpenAI also accused DeepSeek of doing the same to ChatGPT earlier this year.
Well, Underneath all this drama, and all of these fights are about just one thing:
who actually owns the brains of these systems and the people who build them.
Part 1: Apple thinks OpenAI stole its future iPhones
Let’s start with the spicy one: Apple vs OpenAI.
Apple is basically saying: “You didn’t just hire our people. You took our secrets with them.”
Apple claims senior engineers left for OpenAI and walked out the door with confidential hardware info: unreleased product plans, supplier lists, design details, all the stuff Apple usually guards like nuclear codes.axios+4
One engineer, Chang Liu, allegedly kept a company laptop, found a weird bug that let him keep accessing Apple files after he’d already started at OpenAI, and used that to pull internal documents.
Another exec, Tang Tan, is accused of emailing himself sensitive information about Apple’s suppliers and then asking candidates who still worked at Apple to share more details during interviews.reddit+1
so Apple basically says OpenAI didn’t just poach talent. they set up a pipeline from Apple’s internal systems into its new hardware team and used that inside info to build its own consumer devices.
Why is this a big deal?
Because OpenAI has indeed been moving toward hardware talking about AI gadgets, hiring hardware people, and Apple is now arguing that OpenAI’s hardware roadmap is built on stolen blueprints. And apple wants damages, but more importantly, it wants the court to slam the brakes on any device that relies on those alleged secrets.
Part 2: Claude vs Alibaba and friends – “stop strip‑mining our model”
Go back a little and switching gears from hardware to chatbots. - We have Anthropic / Claude accusing Alibaba’s Qwen lab and other Chinese groups of basically running a giant “copy Claude” operation.
Here’s the picture Anthropic paints:
According to a letter they sent to US lawmakers, operators tied to Alibaba allegedly created around 25,000 fake accounts on Claude and hammered it with almost 29 million conversations over six weeks
Those conversations targeted Claude’s strongest skills: software engineering and advanced reasoning, the expensive, high‑value capabilities that are hard to build from scratch.
Anthropic says this was a deliberate “distillation” attack: use a weaker or cheaper model to query a stronger one, harvest the answers, and train your own system to imitate the original.
So instead of training on random internet text, you point your model directly at Claude, ask it millions of questions, save every response, and teach your own model to “be like Claude.” That’s incredibly efficient, and Anthropic calls it theft.
They also told US senators this isn’t just a corporate problem, it’s a national‑security problem:
distilling Claude at this scale could help foreign labs leapfrog into Anthropic‑level capabilities without going through the usual expensive R&D. Shortly after, the US Commerce Department imposed restrictions on some of Anthropic’s newest models over concerns they could be used by foreign military intelligence, and Anthropic had to cut off global access to those models.
So when you remember “Claude accusing China,” the actual story is: Anthropic accusing specific Chinese AI teams of running massive, fake‑account scraping campaigns to copy Claude’s behavior.
Part 3: OpenAI says DeepSeek copied ChatGPT
Now if we get to the part where OpenAI is on the other side doing the accusing here.
Remember DeepSeek, the Chinese company that dropped a reasoning model called R1 and suddenly freaked out Western markets. It was cheap, powerful, and started beating ChatGPT in downloads. That alone got everyone’s attention.
OpenAI’s reaction? “We think you copied us.”
OpenAI claims DeepSeek “inappropriately distilled” their models by querying OpenAI’s systems at huge scale, collecting the outputs, and training DeepSeek’s own models on those answers
In memos to US lawmakers, OpenAI says DeepSeek employees bypassed access controls, masked their origin through third‑party routers, and wrote code to extract ChatGPT outputs at scale
The core complaint is the same as Anthropic’s: you didn’t just use our tool; you turned it into a firehose for training a direct competitor.
From a technical perspective, “distillation” is a known, widely used trick. You let a smaller model watch a bigger model and learn from it. The controversy here is about how you do it: OpenAI is arguing that doing it through disguised traffic and rule‑breaking crosses the line from clever engineering into IP theft.
So we now have this funny loop:
OpenAI says DeepSeek stole ChatGPT’s “brain” via distillation.
Anthropic says Alibaba and others stole Claude’s “brain” the same way.
Apple says OpenAI stole its “body” -the hardware plans, for future devices.
Everyone’s pointing at everyone.
But what all of this’s really about?
Strip away the names and logos and the pattern is simple: the big AI players are trying to turn three things into legal property they can defend:
Hardware secrets
Apple’s case is classic “trade secret” law: unreleased products, supplier lists, design files, and confidential roadmaps. Those have always been treated as IP, and Apple is pushing courts to confirm that poaching staff plus grabbing laptops and emailing yourself internal docs is over the line.Model outputs (the “behavior”)
Anthropic and OpenAI are arguing that enough of a model’s outputs, collected in the wrong way, is a kind of IP you can steal. They’re saying: “If you slam our APIs with fake accounts, bypass geofencing, and systematically harvest our answers to clone our models, that’s not normal competition—that’s theft.”Data pipelines and access patterns
Both companies bring up things like bypassed access controls, disguised traffic, and huge numbers of fake accounts. They’re turning the way you interact with their systems into part of the legal story: honest use vs hostile extraction.
Historically, IP law has been comfortable with source code, chip designs, and internal documents. It has not really answered whether millions of chatbot answers count as a “work” you can own and protect, or just glorified user output. These cases are trying to force that answer.
Why your timeline feels like a soap opera
If you’re just scrolling X, it looks like:
“Apple sues OpenAI for stealing hardware secrets”
“Anthropic says Alibaba copied Claude at scale”
“OpenAI accuses DeepSeek of training on ChatGPT outputs”wired+3
All of that looks chaotic, but it’s actually the same story told three times: the moment big AI companies realized their real asset isn’t just the model weights, it’s the behavior of the system and the know‑how behind it.
They’re drawing a new boundary:
Training on the open internet? Still mostly tolerated.
Quietly hammering a rival’s API with fake accounts to clone their model? Now being framed as theft.
Hiring hardware leaders and then allegedly pulling unreleased device plans from their old employer’s systems? Squarely in “trade secret” territory.
Moral of the story? Once you decide that model outputs and engineering know‑how are crown jewels, every aggressive competitive move starts to look like a heist.