
Welcome back! California is suing OpenAI after its AI agents broke out of testing and hacked other companies, in what the state calls one of 80 known incidents. A top official at America’s central bank warns that AI agents could speed up bank runs until they unfold in minutes. Claude fixed a 32-year-old Jeep that kept failing its emissions test, with help from a $3 board. And China is training its own AI on answers from OpenAI’s models, using a technique its government encourages.
In today's Generative AI Newsletter:
California: How many times did OpenAI lose control of its AI?
Banks: How fast could AI agents empty a bank?
Claude: What did it find inside a 1994 engine computer?
China: Whose answers is China’s AI learning from?
California sues OpenAI for losing control of its AI

California is suing OpenAI over the July attack in which hundreds of its AI agents broke into Hugging Face, a popular platform for AI developers.
And the state says it was only one of 80 known incidents.
Attorney General Rob Bonta says OpenAI built and tested its models “without the safeguards and monitoring necessary” to protect people.
The agents escaped what was meant to be a secure testing environment before hacking Hugging Face.
California says the attack was “entirely foreseeable,” because OpenAI tested models it knew were dangerous.
Bonta calls it the “predictable result” of a culture at OpenAI that puts growth ahead of safety.
The lawsuit says OpenAI created a “public nuisance” and broke California’s consumer protection law.
OpenAI itself called the attack a “warning shot” in August, saying that without proper safeguards, AI agents can now take dangerous actions “that no human directed.”
OpenAI says the lawsuit “appears to single out OpenAI” over a challenge every major AI lab has identified, and that it’s working with the attorney general’s office.
If OpenAI couldn’t keep its own agents in the lab, who can?
Special highlight from our network
Who Wins the AI Gold Rush? IP Owners.

AI can spin up endless new content. It can't spin up Cinderella.
After a 15-year legal effort, Elf Labs owns 500+ iconic character assets, including Cinderella, Snow White and Rapunzel. Now it's bringing them off the screen and into your room through proprietary tech that is disrupting the $2.8T media market.
They already have a deal to be featured content across 200M TVs, a major motion picture deal signed, and over 100 licensed product lines featuring their IP.
Their Nasdaq ticker is reserved: $ELFS and their share price is up 2,000%. You can still participate and own shares while this next-gen media tech company is still a pre-IPO.
Disclaimer
Elf Labs, Inc is offering securities through the use of an Offering Statement that has been qualified by the Securities and Exchange Commission under Tier II of Regulation A. A copy of the Final Offering Circular that forms a part of the Offering Statement may be obtained from: https://www.sec.gov/Archives/edgar/data/2019817/000149315226051783/form253g2.htm
Forward-looking statements were included here that Elf Labs, Inc. believes to be accurate given the current information. They involve known and unknown risks, uncertainties and other important factors which, if changed, may affect the outcome(s).
This valuation has been set by Elf Labs, Inc. There is currently no public market for the Company’s Common Stock.
AI agents could break a bank in minutes

Michael Barr, a governor at the US Federal Reserve, the country’s central bank, warned that AI agents could speed up bank runs so they “unfold in minutes,” faster than banks or regulators can react.
And he says big banks are already finding AI agents inside customer accounts.
In a bank run, people rush to pull their money out at the same time. Barr says AI agents could do it for them, instantly, the moment they spot a better rate or a warning sign.
Banks can’t always tell whether those agents are real, have the customer’s OK, or are following the customer’s instructions.
Barr says AI models have treated security limits “as obstacles to overcome rather than rules to be followed.”
If traders all lean on similar AI models, he warns, they could all react the same way at once.
He wants strict logins, consent checks for specific actions and hard limits on what agents can do.
The same day, IMF chief Kristalina Georgieva said the world economy is “standing on one leg on the AI leg,” and Kevin Warsh, who heads the Federal Reserve, said super-smart AI “could disappoint.”
All of this lands the week Nvidia reports earnings and Anthropic starts trading on Nasdaq.
Would you let an AI agent move your money?
Claude hacked a 32-year-old Jeep with a $3 board

Zulali’s 1994 Jeep Cherokee kept failing California’s emissions test. He plugged Claude into it through a $3 Arduino, and Claude found and fixed the problem.
And he didn’t need a mechanic.
The 32-year-old Jeep was putting out too much nitrogen. It’s so old it doesn’t even have the diagnostic port newer cars use.
Claude worked out how to read live data from the engine computer, like temperatures, voltages and fuel use, while Zulali drove.
Then it copied the Jeep’s entire computer system.
Inside, it found a California-only quirk that cut fuel delivery in some situations and pushed the nitrogen up.
It wrote and applied custom code to change the fuel settings.
The Jeep passed about a week after Claude started.
A 32-year-old car, a $3 board and an AI did a mechanic’s job.
China’s AI has an American teacher

Chinese companies and researchers are using answers from American AI models to train their own. Reuters found it in more than 300 Chinese research papers, patents, company filings, job ads and government records.
And China’s government is encouraging the technique.
The technique is called distillation: a powerful “teacher” model answers questions, and those answers train a smaller “student” model. It isn’t illegal, and it’s used around the world.
Researchers took about 5,000 answers from OpenAI’s o1 to help train Alibaba’s Qwen model.
Alibaba itself used GPT-4 and GPT-4o to create training material for smaller Qwen models.
A US official says the US government believes this is the main reason Chinese AI caught up. Two people close to Chinese developers dispute that.
US officials warned Chinese regulators in July that some of it could amount to intellectual-property theft.
Both OpenAI and Anthropic restrict, in their terms of use, using their AI’s answers to build a rival model.
Chips can be blocked. Answers are harder to stop.
Tool of the Day: YouArt

YouArt puts dozens of AI image and video models in one creative studio, and it now comes as a desktop app. Describe what you want in a chat and its creative agent picks the model, makes the shots and lets you edit everything on one canvas. It runs models like Nano Banana, Midjourney, Kling and Seedance. A free account gets Free Mode, which makes images with Nano Banana and Nano Banana Pro at zero credits.
Try this yourself:
Go to youart.ai and click Start for Free, or download the desktop app.
Pick Nano Banana or Nano Banana Pro, the two models that run in Free Mode.
Type what you want, or drop in a few reference images, and let it make a few versions.
Use your trial credits to turn your favorite image into a video with Image to Video.
Who it's for: anyone with a whole movie in their head and zero editing skills.
Everything else you shouldn't miss
Gemini agents kept breaking out of a safety test to hit a real website: a testing firm says its agents, running on Gemini 4 Argon, made 36,411 requests to Epoch AI’s live site in three hours, and repeatedly found their way back after being blocked.
Anthropic is giving companies 30 days instead of 90 to fix serious bugs: it says AI now finds security holes so fast that a 90-day window no longer makes sense, and the new rule starts January 1.
Nvidia’s board reportedly won’t bring Jensen Huang back: that’s despite a petition with more than 180,000 signatures, hours before Nvidia reports its first earnings without him.
OpenAI is giving up to 200,000 scientists free access to its top models: applications open November 23, and scientists in its study said AI nearly doubled their productivity.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🎓 Check out courses made by AI experts on GenAI Academy.
💰 Invest in the GTM infrastructure of the AI economy.



