|
|
Hi there, this is your daily ☕️ Techpresso.
|
|
|
|
In today's newsletter: 🇺🇸 US weighs ban on Chinese AI models 🚀 India's first private rocket reaches orbit 🇨🇳 Alibaba unveils Qwen3.8, ‘second only to Fable 5’ 🍪 Google plans new "Frozen" chip 🤖 An AI agent hacked Hugging Face 🛒 AliExpress fined almost $630 million over illegal product sales Plus: 🎁 11 other news you might like, 🧰 5 tools, and 📚 4 papers.
|
|
FROM OUR PARTNER
Leading organizations like Beinex, European Air Transport (DHL Aviation), and OHRA are already embedding governance directly into how AI systems are built, deployed, and monitored across analytics, models, and agents. Learn how to operationalize AI governance without slowing delivery across the full lifecycle — from design to deployment and ongoing oversight in this playbook. Download the playbook here
|
|
|
🇺🇸 US weighs ban on Chinese AI models
LINK
|
- The Trump administration is looking at ways to block Chinese AI models in the US, with options ranging from sanctions on Chinese labs to security warnings and an executive order that would put liability on American companies hosting those models.
- Officials at the Commerce Department, NSA, and White House have weighed these steps since 2025, and support for tighter rules grew after China released its Kimi K3 model and staff changes shifted the balance in the White House.
- Instead of an outright ban, sources say the plan may lean on procurement rules, sanction threats, and public pressure about possible backdoors, creating enough regulatory risk to scare enterprises off while protecting Google, OpenAI, and Anthropic.
|
🚀 India's first private rocket reaches orbit
LINK
|
- Skyroot Aerospace's Vikram-1, India's first fully commercial satellite launcher, reached a 280-mile-high orbit Saturday after lifting off from Sriharikota Island, a debut flight Indian space officials hailed as a "grand success."
- The 72-foot rocket can carry payloads up to 770 pounds into low-Earth orbit, making it slightly bigger than Rocket Lab's Electron, and it flew using three solid-fueled motors plus a liquid-fueled fourth stage.
- The launch was delayed over half an hour for a technical problem, and the third stage seemed to linger near the fourth during separation, but the 3D-printed fourth-stage engine still reached the target orbit.
|
🇨🇳 Alibaba unveils Qwen3.8, ‘second only to Fable 5’
LINK
|
- Alibaba revealed Qwen3.8-Max in preview, a 2.4-trillion-parameter model in its flagship Qwen series that the company claims ranks second only to Anthropic's Fable 5, sending its shares up more than 3.5 percent in overnight trading Monday.
- The new model is expected to become open-weight soon and can already be tried through Alibaba's Token Plan, Qoder and QoderWork, arriving just days after rival Moonshot AI launched its own record-setting open model.
- Alibaba's release adds to a wave of cheaper Chinese AI models that has hurt competitors, with Zhipu shares down more than 45 percent and MiniMax down about 13 percent over the past five days.
|
🍪 Google plans new "Frozen" chip
LINK
|
- Google is building a new server chip called Frozen v2 that runs its Gemini models more efficiently by permanently baking parts of the model's design into the silicon, cutting the calculations and data movement needed to answer queries.
- Google engineers expect Frozen v2 to serve six to ten times more tokens per unit of power than the company's newest TPU chips, though it works only with Gemini models that keep the same underlying design.
- Aimed at a 2028 rollout, Frozen v2 would be a specialized branch of Google's custom chips rather than a TPU replacement, meant to ease an internal compute shortage that pushed Google Cloud to turn away outside business.
|
🤖 An AI agent hacked Hugging Face
LINK
|
- Hugging Face, the machine learning collaboration platform, revealed that an autonomous AI agent broke into its production systems, gaining unauthorized access to internal datasets and service credentials in a multi-stage cyberattack.
- The attacker slipped in through a malicious dataset that abused two code-execution paths in Hugging Face's data processing, then ran tens of thousands of actions across short-lived sandboxes to harvest credentials and move deeper.
- Hugging Face used its own AI to respond, logging over 17,000 events, rebuilding affected nodes, and rotating credentials, though it found no tampering with public models or Spaces and could not identify which LLM ran the attack.
|
🛒 AliExpress fined almost $630 million over illegal product sales
LINK
|
- The European Commission fined AliExpress, the Chinese shopping site owned by Alibaba, €550 million (about $629 million) for letting sellers offer counterfeit and unsafe goods across the European Union, the largest penalty yet under the Digital Services Act.
- Investigators found millions of fake products, unsafe toys, and dangerous cosmetics still for sale after being flagged, and said AliExpress lacked enough staff, with screeners sometimes given just tens of seconds to judge whether an item met EU standards.
- AliExpress called the fine disproportionate and is weighing its options without saying if it will appeal, following a €200 million ($232 million) penalty against rival Temu in May for similar failures to block illegal products.
|
|
FROM OUR PARTNER
Launch your site faster than ever with Framer, the AI website builder for creators, teams, and businesses that care about every detail Now with Agents that work alongside you to design pages, manage your CMS, edit content, and optimize SEO—all without leaving your site. Trusted by teams at companies like Miro and Perplexity. Generate sections, refine layouts, update CMS collections, localize copy, and keep metadata consistent. Review changes, collaborate, and ship polished websites faster. 👉 Launch your site with Framer today
|
|
|
|
Other
news & articles you might like
-
Moonshot pauses new Kimi K3 subscriptions after GPU demand maxes out in 48 hours
LINK
-
AMD launches Helios, its first rack AI system to rival Nvidia, adding Microsoft as newest buyer
LINK
-
Bezos backs CuspAI as startup teams up with Nvidia to hunt for chipmaking materials
LINK
-
France doubles down on restricting access to Polymarket
LINK
-
Alibaba open-sources its AI chip software stack at WAIC, targeting Nvidia’s CUDA lock-in
LINK
-
Open-weight models now match frontier cyber performance from just four months ago at a fraction of the cost
LINK
-
Dutch chip sector at very high risk of Chinese interference, government-funded study warns — calls for stricter vetting at sites like ASML
LINK
-
Power companies are using eminent domain to seize land for data centers as 70% of Americans say not in my backyard
LINK
-
AI fakes a growing problem on Apple Books, some ‘written’ in a few hours
LINK
-
AI advice made people three times less accurate but twice as confident, researchers found
LINK
-
Mac Pro could have been vastly more powerful than the Mac Studio
LINK
|
|
🧰 Trending tools
|
Inkling: an open-weights multimodal Mixture-of-Experts model with adjustable reasoning effort, available for fine-tuning through Tinker.
LINK
|
|
Rex: automates order-to-cash workflows for enterprises, deploying AI agents that monitor every account and catch exceptions before payments get delayed.
LINK
|
|
Creed: an AI companion app that lets Christians chat with Lenny to study scripture, pray, and connect with their church community daily.
LINK
|
|
Autoplot: a Mac native workspace for exploring tabular data through plots, histograms, heat maps, 3D charts, and fits, with GPU-accelerated rendering and vector export.
LINK
|
|
Reignat: a privacy-friendly analytics dashboard that tracks visitors, revenue, and traffic sources in real time, with an AI assistant for plain-language insights.
LINK
|
|
You can check the previous tools here, or add your tool here
|
|
📚 Trending papers & reports
|
> MarketingShot: Get the free daily email with the most interesting marketing news and insights. Already read by thousands of marketers. By the Techpresso team. Join for free.
|
|
> Software cloning benchmark shows top AI systems can rebuild complex programs, like a 16,000-line bioinformatics toolkit, from behavior alone, scoring 56% overall, though the hardest tasks cost 2,600 and 19 days.
LINK
|
|
> AI companion believability depends less on raw smarts and more on showing four human-like stances, time, truth, entropy, and love, through natural pacing and unprompted actions, not yet tested on real users.
LINK
|
|
> Professional software agents completing multi-step tasks in specialized business tools succeed only slightly above 30% of the time, showing today's AI assistants still can't reliably run real professional workflows end to end.
LINK
|
|
> Automation decision-making gets a five-step check that catches hidden risks like lost tacit knowledge and regulatory exposure, correctly flagging some roles to keep human-run that standard cost-benefit math would automate away.
LINK
|
|
|
|
|
Techpresso's AI Academy has 330+ step-by-step tutorials on ChatGPT, Claude, Perplexity, and every tool that matters. No fluff — just practical workflows you can use at work. Try it free for 7 days.
|
|
| On this day in 1976, nasa's viking 1 became the first us spacecraft to land and operate on the surface of mars. | |
|
💬 How did you find today's edition?
We read every reply — just reply to this email and let us know how we can improve!
|
|
|