The Tesla Cybercab is finally on sale to regular buyers — but the car won’t move unless you keep paying every month. Meanwhile, a forensic study claims to have found the fingerprints of export-banned Nvidia chips inside some of China’s most advanced AI models, just days after Nvidia’s CEO resigned. Here’s everything that mattered this weekend.
Key Takeaways
- Tesla Cybercab costs $29,990, but a mandatory $299-a-month subscription is required to drive it — and only inside Tesla’s Robotaxi areas.
- Six Chinese AI models carry traces of Nvidia Blackwell chips, which U.S. export rules bar from China — though the evidence doesn’t prove where the chips were or whether any law was broken.
- Ransomware groups are using AI for better phishing emails and ransom negotiations, with claimed attacks up sharply this summer.
- DeepSeek has opened the search, file-reading, and voice tools from its app to developers through its API.
- Anthropic’s Claude can now join Google Meet calls as a visible participant, taking notes and drafting documents live.
1. Tesla Cybercab: $29,990 to Buy, $299 a Month to Drive
Tesla has started inviting select customers to buy a Cybercab for personal use, and the headline price is exactly the one Elon Musk promised two years ago: $29,990 before taxes. The catch, as The Verge reports, is sitting right next to the price on the order page: a $299-a-month Autonomy Subscription, required to operate the car. The first month is included, a $1,000 deposit is non-refundable, and the fine print is blunt — “Subscription and regular vehicle maintenance required to operate.” Cancel the subscription and the car simply won’t drive on public roads.
The Cybercab only works inside Tesla’s Robotaxi service areas — eight cities in Texas and Florida — and buyers can’t use it commercially, so there’s no way to earn fares to offset the fee. Musk himself called the $299 fee “too high” and said it should be closer to $115 a month, but the order page still lists $299, and Tesla can raise it with 30 days’ notice. First deliveries are estimated for December 2026 in Austin.
Do the math and the “affordable robotaxi” looks less affordable: $299 a month adds nearly $17,600 over five years — about 60% of the sticker price. Still, it’s a milestone: the first driverless car built without any manual controls sold to private buyers.
2. The Blackwell Fingerprint: Did Chinese AI Models Train on Banned Nvidia Chips?
As TechCrunch reports, a nonprofit called Unemployment Labs — led by former OpenAI researcher Will DePue — published a forensic analysis claiming six open-weight AI models from four Chinese labs (DeepSeek, Moonshot AI, MiniMax, and Ant Group) carry traces of Nvidia Blackwell GPUs in their published weights. Blackwell is Nvidia’s current chip generation, and U.S. export rules require a license to sell it to China. For DeepSeek-V4-Flash, the evidence is called conclusive.
But the caveats matter. The fingerprint identifies the hardware that created the models’ starting weights — it supports, but does not prove, that training ran on the same chips — and says nothing about where the GPUs were, how access was obtained, or whether any rules were broken. Renting Blackwell capacity from data centers outside China is largely legal.
The timing is hard to ignore: the report landed days after Jensen Huang resigned as Nvidia CEO amid a chip-smuggling scandal — a story we covered in our November 13 roundup. Lawmakers are now pushing for stricter export curbs, and the study gives them something new: evidence anyone can rerun from public data.
3. AI Is Making Ransomware Nastier — Just Not in the Way You’d Expect
Ransomware is having a boom year, and AI is helping — but not by writing unstoppable malware. According to WIRED’s reporting, language models are making the human parts of the attack cheaper and more convincing: better phishing emails, localized scams, target research, and even ransom negotiations. Researchers tracked nearly 3,000 claimed incidents over the summer, with one analysis putting the quarterly rise at around 30%.
The reassuring part: fully AI-powered ransomware hasn’t emerged yet. As Recorded Future researcher Allan Liska told WIRED, the successful groups are still largely human-operated — of 405 AI-linked malware samples tracked by Halcyon this year, only 12 reached a protected endpoint, and all 12 were detected. The real risk is volume: AI lets small crews punch above their weight. The defense hasn’t changed much — strong passwords, multi-factor authentication, and patched systems still stop most of these attacks — but the attackers’ front door is getting a lot more polished.
4. DeepSeek Hands Developers the Tools From Its Own App
DeepSeek has opened up the toolkit behind its consumer chatbot to developers. As Crypto Briefing reports, the Chinese AI lab announced on November 13 that the search, file-reading, and voice features from the DeepSeek app are now available through its API. Developers can add web search with citations, URL reading, document parsing for PDFs and Office files, content safety checks, and voice features to their own products, all as simple switches inside existing API requests.
There are also standalone APIs for search, page fetching, OCR, speech-to-text, and text-to-speech, each billed separately from token prices, with the voice tools currently half price. OpenAI and Google have offered many of these capabilities for a while, but DeepSeek’s pricing undercuts them sharply. The new deepseek-flash model powers it all, and DeepSeek released an open-source demo chat client too.
5. Claude Can Now Sit in Your Google Meet
Anthropic has started rolling out Claude Meetings in beta, a feature that puts Claude inside your video calls as a visible participant. According to Anthropic’s announcement, Claude joins Google Meet calls, follows the conversation, answers questions in the chat, and starts drafting documents and whiteboards while the meeting is still happening.
It’s available to Team and Enterprise plans only — no Pro or Free tiers yet — and Google Meet is the only supported platform for now. Notes and transcripts are free during the beta, and documents and whiteboards created in calls are free through February 10, 2027. There are sensible guardrails: Claude stays out of calls with outside guests by default on Enterprise plans, keeps a transcript and summary but not audio, and leaves after up to eight hours. It’s a direct challenge to the note-taking bots already in everyone’s meetings — except this one is supposed to do the work during the call, not just summarize it afterwards.
The Takeaway
This weekend’s stories share a theme: the AI era is moving from demos to dollars — and to consequences. Tesla is selling a car you rent the right to drive, DeepSeek is selling the tools from its app, and a forensic report is turning open model weights into export-control evidence. The technology keeps getting more capable; the interesting part is increasingly what it costs, who controls it, and who gets to check the receipts.
Check back tomorrow for the next roundup.
Frequently Asked Questions
How much does the Tesla Cybercab cost in total?
The Cybercab is listed at $29,990 before taxes, plus a $1,000 non-refundable deposit and a mandatory $299-a-month Autonomy Subscription. Over five years, the subscription alone adds about $17,600, making the real cost closer to $47,600 before taxes, insurance, and charging.
What is the Blackwell fingerprint in Chinese AI models?
It’s a forensic finding from Unemployment Labs: numerical patterns in the published weights of six Chinese AI models match the thread layout of Nvidia’s Blackwell GPUs, which U.S. export rules bar from China. The evidence shows the hardware that created the models’ starting weights, but not where the chips were or whether any law was broken.
Is AI making ransomware attacks worse?
Yes, mainly by making attacks cheaper and more convincing. Ransomware groups use language models for phishing emails, target research, and ransom negotiations, and claimed attacks rose sharply this summer. Fully autonomous AI ransomware hasn’t emerged, though, and most attacks still rely on stolen credentials and human operators.
What is Claude Meetings?
Claude Meetings is Anthropic’s beta feature that adds Claude as a visible participant in Google Meet calls. It takes notes, answers questions, and drafts documents and whiteboards live during the meeting. It’s available on Team and Enterprise plans, with Google Meet only at launch.
Sources
- The Verge, “Tesla is inviting select customers to buy a Cybercab for personal use at $29,990 plus a $299-a-month autonomy subscription” — https://evselect.com.au/articles/tesla-cybercab-personal-orders-29990
- TechCrunch, “Chinese frontier AI models carry the fingerprint of export-controlled GPUs” (via Unemployment Labs) — https://intelligibberish.com/articles/2026-11-14-fingerprinting-gpu-inside-open-weight-models/
- WIRED, “Beware the New AI-Fueled Ransomware Boom” (Kernel Panic) — https://eletiofe.com/beware-the-new-ai-fueled-ransomware-boom/
- Crypto Briefing, “DeepSeek opens its full app toolkit to developers through its API” — https://cryptobriefing.com/deepseek-opens-app-features-developers-api/
- Anthropic, “Introducing Claude Meetings” — https://vanja.io/claude-meetings/