Tokens by Michael Brenndoerfer, Week #4
Apple Sues OpenAI Over Stolen Blueprints and Illegal Show-and-Tell
Today's issue: Apple takes OpenAI to court over hardware smuggling, Spain targets unsupervised kids on LLMs, and Waymo snitches on its own riders.
Welcome to #4 of your weekly AI newsletter.
Don't miss the next one 📬
Everything that mattered in AI this week, in five minutes. Occasionally hilarious.
No spam. Unsubscribe anytime.
Main Story
The corporate show-and-tell is all fun and games until the lawyers show up.
The Apple-OpenAI Honeymoon Ends in Federal Court
The tech partnership of the year is officially dead, buried under a federal lawsuit. Apple is suing OpenAI and its hardware partner, io Products, alleging systemic theft of trade secrets. The complaint claims OpenAI poached former Apple employees and coached them to bypass exit protocols while leaking battery designs, logistics files, and manufacturing blueprints.
The details are wild. Apple accuses OpenAI Chief Hardware Officer Tang Yew Tan of instructing job applicants to bring physical Apple parts to their interviews for an illicit show-and-tell. Meanwhile, another former employee allegedly used an authentication bug to breach Apple's internal network and download confidential files right before resigning.
This lawsuit ends their brief era of cozy collaboration. It is likely no coincidence that the filing lands right after Apple decided to power its revamped Siri using Google Gemini instead of ChatGPT. When the honeymoon ends in Silicon Valley, it ends with subpoenas.
- The show and tell: OpenAI hardware chief Tang Yew Tan allegedly had applicants bring physical Apple parts to interviews.
- The network breach: Former employee Chang Liu used an authentication bug to download confidential Apple files before departing.
- The Gemini pivot: Apple recently opted to power its revamped Siri with Google Gemini, leaving OpenAI in the cold.
Bottom Line: This lawsuit turns OpenAI's hardware ambitions from a quiet stealth project into a public legal target. If the allegations hold up, the race to build physical AI devices is driving companies to skip standard corporate poaching and go straight for physical hardware smuggling.
Source: www.theguardian.com
Don't miss the next one 📬
Everything that mattered in AI this week, in five minutes. Occasionally hilarious.
No spam. Unsubscribe anytime.
The Feed
-
OpenAI dropped GPT-5.6 after a government review. The cheap Luna model slashes standard API costs by 25x, making high-volume enterprise automation far cheaper.
-
Stanford's new report warns that the AI labor transition is moving far too fast for traditional safety nets.
-
The 18MB Rust binary Mesh-LLM daisy-chains devices over peer-to-peer networks. You can run a massive 235-billion-parameter model on consumer GPUs without cloud providers.
-
AI coding assistants are flooding unpaid open-source maintainers with buggy code. A Financial Times report details projects shutting down bug bounties and banning AI code.
-
Meta pulled its Instagram tool, Muse, after a user revolt. The New York Times reports that opting in public accounts by default sparked immediate privacy concerns.
-
Reuters found that Meta's image detector fails 55 percent of the time when an AI image is simply cropped or screenshotted. Standard watermarks cannot stop basic editing.
-
Anthropic is putting Fable 5 on a pay-as-you-go model. The new pricing structure means autonomous coding loops will run up massive bills instead of staying flat-rate.
-
Want to train better AI agents? Submerge them in office chaos. Bespoke Labs raised $40 million to build testing environments using broken databases and messy Slack threads.
-
In his new book, computer scientist Peter Denning argues that we cannot math our way to common sense. ScienceDaily reviews the argument that AI requires a physical body to comprehend our world.
-
The open-source Ghost Font blocks AI vision models and OCR tools while remaining legible to human eyes. It protects sites from scrapers without annoying readers.
-
Spain proposed a UN coalition to protect children from unsupervised LLMs.
-
Fujitsu's new AI modernization service blends Claude, GPT, and custom LLMs to speed up legacy mainframe migrations by 40 percent.
-
A Waymo vehicle called the police on its own teenage passengers after detecting drinking and a toy gel-blaster. Your driverless taxi is a snitch.
-
China's foreign ministry endorsed open-source AI to help enterprises slash API costs and challenge closed Western models. This signals a geopolitical push for open weights.
-
The AMA highlighted a new Senate bill to study the safety of AI-driven voice companions and clinical diagnostics marketed to seniors.
See you next week!
Michael
Get Tokens in your inbox 📬
Everything that mattered in AI this week, in five minutes. Occasionally hilarious.
Join 5,000+ readers · No spam, unsubscribe anytime.