# OpenAI Just Dragged Apple Into the Dirt Over Those Trade Secret > Source: > Published: 2026-08-05 03:44:17+00:00 # OpenAI Just Dragged Apple Into the Dirt Over Those Trade Secret **Apple is getting this wrong**, and OpenAI is not here for it. Apple sued OpenAI last month, claiming two ex-engineers—Chang Liu and Tang Tan—blew the whistle on Apple’s confidential hardware schematics and stole trade secrets to help OpenAI build its own consumer gadgets. Apple even asked a judge to block those guys from touching any Apple IP. Meanwhile, OpenAI dropped a 3,000-word takedown on its blog titled * Apple is getting this wrong*, backed by screenshots of Slack DMs, emails, and timestamps. Here’s the tea: - Apple’s own outside counsel confused Chang Liu with another Asian employee during initial outreach, then admitted they never actually spoke to OpenAI’s legal team before filing suit. Crickets for five months afterward. - Chat logs show *Apple employees*reaching out to Liu asking for help debugging firmware—not the other way around. So much for the rogue defector narrative. - The “residual access” Apple claims? Turns out it was just sloppy offboarding procedures on their end. Classic. OpenAI concluded: no secrets, no theft, no problem. Case closed—or at least thrown into the dumpster where it belongs. WeChat's “Strangers Mode” Exists Already—And It's Weirdly Useful Turns out WeChat *did* ship something like temporary chats—but quietly, buried under layers of menus like an Easter egg only grandmas know how to find. Forget adding someone as a friend. Want to share photos or files? Scan someone’s QR code via `Scan → My Code → Face-to-Face Transfer` . Need to send cash without saving contacts? Use `Face-to-Face Red Packet` or tap into `Payments` > `Transfer` to dump money straight into a bank account or phone number. Need to make a group chat fast? Go to `Add Friends` > `Face-to-Face Group` , set a 4-digit PIN, invite everyone nearby, then dissolve the chat once the meetup ends. No trace left behind. It’s clunky, yes—but functional. And honestly, more useful than TikTok’s new AI friend feature that nobody asked for. Huawei Responds to Bamboo Rattle Drama: Not Banning Toy Sales A viral meme comparing a bamboo rattle toy to Huawei exec Richard Yu’s keynote cadence sparked a firestorm online—and now Huawei is stepping in. In a statement posted late last night, Huawei denied any involvement in banning or removing the product from e-commerce platforms. They clarified that ongoing IP crackdowns target specific infringing content—not the toy itself—and reaffirmed their stance on protecting legitimate speech. They also confirmed they haven’t contacted Taobao, JD, or any retailer to pull listings. The rattle stays live. Crisis averted—for now. VW’s Meltdown Continues: Layoffs, Factory Closures, and a 25% Hit in China Volkswagen just announced its biggest restructuring ever—and it looks grim. By 2030, VW plans to slash global production capacity by 25%, cut car variants by half, reduce parts inventory by 75%, and lay off 50,000 indirect staff. Sales and admin costs will drop from 16% of revenue to 12%. ROA jumps to 8–10%. Sales? Down 6.3% globally, 25.9% in China alone. Profits fell 11.6%. Revenue growth forecast revised from +3% to flat. Yikes. Meanwhile, VW’s Chinese arm is doubling down on local partnerships—with XPeng, Horizon Robotics, and others—to speed up EV development timelines. Time will tell if these moves save them—or sink them deeper. China Drops Mandatory Safety Standards for L3/L4 Autonomous Driving Systems The Ministry of Industry and Information Technology rolled out new regulations today requiring automotive manufacturers to meet strict safety benchmarks for Level 3 and Level 4 autonomous driving systems starting July 2027. Named GB/T 44721-2026, these mandatory standards cover design validation, simulation testing, road trials, cybersecurity protocols, driver monitoring capabilities, and fail-safe mechanisms across the entire lifecycle—from concept to decommission. Notably excluded: automated parking assist functions. Those remain under separate oversight frameworks. Automakers must now implement robust internal quality control systems, maintain detailed audit trails, and undergo third-party certification processes before deploying Level 3+ features in production vehicles sold within mainland China. This move signals growing confidence in domestic autonomous tech maturity—and raises questions about whether foreign OEMs can keep pace with increasingly stringent compliance requirements. Tencent Extends Free Access to Hunyuan-Thinking Model Through August 31 Good news for devs who love working with large language models without spending cash: Tencent WorkBuddy announced an extension of its free tier access to the Hunyuan-Thinking model through August 31st. Originally launched earlier this year, Hunyuan-Thinking focuses on advanced reasoning tasks including code generation, document summarization, data analysis workflows, and general-purpose problem solving scenarios requiring deep contextual understanding. While extended usage beyond daily quotas may trigger wait times depending on server load conditions, peak hours (~11 PM–8 AM CST) typically offer smoother performance due to reduced traffic volumes. Tesla Owners Report Overheating Issues With New FSD v14 Lite Update Owners of Tesla vehicles equipped with Hardware 3 (HW3) units have reported frequent overheating warnings following rollout of the latest Full Self-Driving (v14 Lite) update—a stripped-down variant adapted from the original HW4-based version. Despite being optimized for lower memory bandwidth constraints (8GB RAM vs 64GB), some users experienced thermal throttling severe enough to require physical replacement of onboard autopilot computers altogether. At least four confirmed cases so far involve complete hardware swaps performed at authorized service centers nationwide. Tesla support teams are reportedly investigating root causes amid rising concerns over reliability implications associated with pushing AI-heavy updates onto aging platforms lacking adequate cooling infrastructure. SpaceX Teams Up With NVIDIA to Launch Orbital Supercomputer Powered By Rubin GPUs SpaceX revealed plans to launch a constellation of orbital supercomputers powered by next-gen NVIDIA Rubin architecture processors designed specifically for high-performance computing applications in space environments. Dubbed “Starmind AI1,” each satellite hosts dual Rubin GPU accelerators alongside custom-built Vera CPUs engineered to withstand radiation exposure while maintaining teraflops-level compute throughput necessary for training cutting-edge neural networks directly aboard orbiting data centers. Initial deployment targets low Earth orbit (LEO) altitudes ranging between 550km–1,200km above sea level with laser inter-satellite links enabling ultra-low latency communication speeds comparable to terrestrial fiber optic networks spanning thousands of kilometers apart globally. Planned fleet size? One million nodes total—yes, really. That would dwarf current global internet constellations combined and potentially redefine how we think about distributed cloud computing architectures moving forward. [DeepSeek](/en/tags/deepseek/) Dominates OpenRouter Rankings After Processing 7.2 Trillion Tokens Weekly According to latest industry reports compiled by API aggregator OpenRouter, DeepSeek’s V4 Flash model processed over 7.2 trillion tokens during the week ending August 2nd—surpassing all competitors including Google Gemini, Anthropic [Claude](/en/tags/claude/), and even Meta Llama series entries. Breakdown reveals two variants collectively handled 670B tokens independently, contributing significantly toward platform-wide throughput reaching 3.29 quadrillion tokens per second average rate sustained throughout monitored period. Interestingly, several emerging Chinese startups also cracked top ten listings simultaneously—including MiMo-V2.5, Hy3, and Zhipu AI’s GLM-5.2—all leveraging hybrid pricing strategies ranging from completely free offerings to minimal-cost subscription tiers encouraging rapid adoption among early-stage developers seeking scalable alternatives to expensive proprietary toolchains dominating traditional enterprise markets today. Open Source Alternative Emerges From Shadows As DeepSeek Hits 8 Trillion Tokens Daily A lesser-known open-source initiative codenamed OpenCode recently disclosed unofficial metrics indicating that DeepSeek V4 Flash achieved staggering peaks exceeding 8 trillion tokens processed within single calendar days—with nearly two-thirds attributed to organic developer activity originating from community-driven projects rather than commercial deployments. Breakdown suggests approximately 3 trillion tokens came from unpaid experimentation cycles fueled primarily by academic researchers, indie hackers, and independent contributors actively building next-generation AI-powered applications using freely available public repositories hosted across GitHub and Hugging Face ecosystems. Meanwhile, remaining volume traced back to paid enterprise integrations spanning customer service bots, automated content moderation pipelines, and intelligent tutoring systems deployed across educational institutions worldwide. Implication? Community enthusiasm continues fueling exponential growth trajectories previously reserved exclusively for well-funded corporations dominating mainstream discourse around artificial general intelligence advancements lately. [OpenAI Settles $3. 1h ago](/en/news/5038/) [Apple's latest legal filing reveals a concerning pattern 10h ago](/en/news/4966/) [OpenAI Fires Back at Apple's Trade Secret Lawsuit with Public 11h ago](/en/news/4958/) [AGs Order OpenAI to Preserve Hugging Face Hack Records 17h ago](/en/news/4923/) [The Race to Beat Cheap AI from China: What It Really Takes 20h ago](/en/news/4909/) [15 Attorneys General vs OpenAI: The Regulatory Push 23h ago](/en/news/4894/) [Next Why the New Leapmotor Z9S Feels Overpriced Despite Its Tech →](/en/news/5046/) ## All Replies （0） No replies yet — be the first!