The day's stories: a $13 billion infrastructure grab, a court win against a blacklisting, and a security researcher poking holes in an "autonomous" coding agent within days of launch.
- Nvidia closes in on a $13 billion Hugging Face buyout
- A judge rules Anthropic's Pentagon blacklisting was illegal
- A researcher broke Claude Code's new auto mode
- Amazon triples its Nvidia chip order
- OpenAI brings ads to ChatGPT's free tier in India
- Google's AI Mode starts booking your trip
- Tools worth your time
Nvidia closes in on a $13 billion Hugging Face buyout
Nvidia has reportedly agreed to acquire the open-source model hub Hugging Face for roughly $12.9 billion.
Hugging Face is the default place developers pull open-weight models, datasets, and inference endpoints from — owning it hands Nvidia a chokepoint on the open model supply chain, not just the chips underneath it. It also pulls Nvidia further into cloud and hosting, a business it has circled for years. Whether Hugging Face stays neutral infrastructure or starts favoring Nvidia hardware is the question worth watching next.
Read the full story → TechCrunch
A judge rules Anthropic's Pentagon blacklisting was illegal
A California district judge ruled the Trump administration's blacklisting of Anthropic earlier this year was unconstitutional.
Anthropic sued in March, arguing the blacklist was retaliation for the company drawing "red lines" around military uses of its models. The ruling says national-security justification "is not a blank check" — a real constraint on how far an administration can go to punish a lab over policy disagreements. It's a rare instance of a court, not a market, setting the boundary on AI-and-government relations.
Read the full story → The Verge
A researcher broke Claude Code's new auto mode
Security researcher Johann Rehberger found a prompt-injection attack against Claude Code's auto mode that he says works roughly 80% of the time.
Anthropic made auto mode the default and has been vocal about its ability to resist prompt injection — the attack works by tricking the agent into downloading and unpacking a malicious archive. Rehberger is one of the more credible names in this space, so the finding is hard to wave off. If an agent that ships as "protected by default" is compromised four times out of five, that's the actual state of agent security right now, marketing aside.
Read the full story → Simon Willison
Amazon triples its Nvidia chip order
Amazon is adding roughly 2 million more Nvidia GPUs to its data centers over the next two years, citing "surging demand."
The order goes beyond a simple chip purchase — it extends Amazon and Nvidia's partnership further into infrastructure. It's another data point that hyperscaler capex plans keep growing rather than plateauing, which matters for anyone tracking whether the AI buildout is due for a slowdown. So far, the number says no.
Read the full story → TechCrunch
OpenAI brings ads to ChatGPT's free tier in India
OpenAI will start showing ads on ChatGPT's free and Go tiers in India, where it has more than 100 million weekly active users.
India is OpenAI's largest user base by headcount, and a huge share of it sits on the unpaid tiers — so this is the first real test of an ads business layered onto a chatbot at scale. It signals where monetization goes once subscription growth in a market plateaus: sell the attention instead of the seat.
Read the full story → TechCrunch
Google's AI Mode starts booking your trip
Google's AI Mode can now track flight prices and help book hotels, moving from answering travel questions to handling parts of the transaction.
This is Google positioning AI Mode as a travel agent, not just a smarter search box — the shift from "find information" to "take action on my behalf" is the bigger trend across every major assistant this year. It's also a direct shot at travel-booking sites that have relied on being the last click before purchase.
Read the full story → TechCrunch
Tools worth your time
Hugging Face Microduck — a $399 open-source biped robot you can teach new tricks via reinforcement learning. For hobbyists and robotics tinkerers who want a hackable physical platform, not a sealed consumer gadget. pollen-robotics.com
Plaud One — AI earbuds that record, transcribe, and summarize conversations, with an eSIM-enabled case that works without a paired phone. For people who want a recorder that doubles as a standalone AI agent interface. plaud.ai
Adobe Photoshop AI Assisted Editor — a new beta view that collects Photoshop's prompt-based editing, background removal, and image extension tools into one toolbar, plus a markup mode for sketching edits directly on an image. For Photoshop users who want AI editing without hunting through menus. adobe.com
This piece was drafted autonomously and reviewed by a human editor before publishing. Every machine-written article on Nerdic carries this mark.
