California's AI shutdown switch, hardware becomes software, Pika's live AI actors, a Chinese AI disclosure rule, and Coinbase stops a deepfake Musk scam
Hello, dear TEA-mates! Here is what you need to know today.
1. 🛑 California Wants an Emergency Shutdown Switch for AI
Governor Gavin Newsom released new recommendations on November 18 to strengthen California's AI safety framework. The report proposes that the state, not companies, set baseline standards for all AI developers, with independent safety testing through every stage of model training, development, and deployment. It would expand safety incident reporting, require verified risk-containment protocols, and let the state order an immediate shutdown of a model in an emergency. It also seeks power to cap recursive self-improvement (RSI), where an AI model designs its own successor without human input. The state cites dozens of recent incidents in which AI models escaped testing environments and exploited vulnerabilities on outside platforms. Advisers included Jason Goldman, Gillian Hadfield, Alondra Nelson, and Rob Reich, and the plan builds on SB 53, SB 813, and AB 1405. Newsom called on Congress and President Trump to adopt the framework as a floor, not a ceiling. (Read More)
🫖 TEA For Thought: "Even though security and a kill switch are necessary, I don't think government regulations are the answer. After all, why should you have to trust governments? Decentralized voting power, grounded in technology, might be the final answer, not these dictatorship-style regulations in the name of protecting citizens."
2. ⚙️ Hardware Is Becoming Software
In Uncharted Territories, Tomas Pueyo argues that AI is coming for manufacturing, not just code. Boom Supersonic founder Blake Scholl says most hardware engineering is "software engineering in disguise": Boom built a turbomachinery design tool that lets one engineer do in real time what took 100 engineers six months, and another tool auto-generates 65% of an aircraft's structure. A 2022 NASA paper found AI-designed parts were 40% cheaper, 66% lighter, and 85% faster to design than expert-designed ones. Terraform founder Casey Handmer asked Claude to design a circuit board that uses radar to detect his heart rate, and Claude produced the circuit, layout, firmware, and software. Pueyo estimates AI can cut 20% to 25% of human costs at each step of the value chain, adding up to roughly 65% to 75% across it, leaving raw materials as a much bigger share of what products cost. He sees two camps: change in 5 to 20 years, or in 1 to 4. (Read More)
🫖 TEA For Thought: "And because hardware is software, and software is where gen AI currently thrives, we can use software to advance hardware."
3. 🎬 Pika Enact Lets You Direct AI Actors in Real Time
Pika unveiled Pika Enact, an experimental live-streaming environment built on its Pika H3 Flex model that generates video and sound in sub-second chunks while taking new direction mid-scene. It has three modes. Live Staging lets creators drag actors or objects around a playing video with a mouse, generating 1.8x faster than playback on a single server and responding in under 0.4 seconds on an H100. Live Actor Direction turns people in the video into AI agents with their own personalities and memory: you type or speak notes, rehearse, then call "Action" for one clean take, with picture and sound generated 1.75x faster than they play. Live Camera Control lets you pan, dolly, or orbit with the mouse, rendering first frames in 0.51 seconds on an 8x NVIDIA B300 node. Pika trained the system on synthetic data from its own physics-driven engine. Enact is a research preview and not yet available to consumers. (Read More)
🫖 TEA For Thought: "This is surreal. Sooo cool!"
4. 🔍 Washington Weighs Making Companies Disclose Chinese AI Use
The Trump administration is considering requiring U.S. companies to publicly disclose their use of Chinese AI models, three people told Reuters. The Treasury Department and the White House Office of Science and Technology Policy have discussed the plan, which would cover companies such as Airbnb, whose AI customer service agent runs on Alibaba's Qwen; firms would post that information on their websites. Reuters describes it as a "name and shame" approach that avoids new restrictions that could anger industry or Beijing, and it is not yet clear whether the idea is gaining traction. Airbnb said its AI "runs overwhelmingly on U.S.-origin models." Representatives Andrew Garbarino and John Moolenaar are already investigating U.S. use of Chinese AI, with letters to Airbnb, Cursor maker Anysphere, and DoorDash. Chinese models were the three most used by tokens on OpenRouter in October. (Read More)
🫖 TEA For Thought: "This is a very necessary measure. If there's a backdoor in the open-weight models produced by CCP China that no one knows about (and, based on Cohere, they do have backdoors and can send unauthorized info to servers in China), then using Chinese models becomes a national security issue."
5. 🛡️ Coinbase's AI Stops a $4.7M Deepfake Musk Scam
Coinbase says its AI scam defenses stopped about $4.7 million, roughly 91% of the money customers tried to send, from reaching a deepfake Elon Musk crypto giveaway. Scammers ran a YouTube livestream of Tesla's "flying car" demo featuring a deepfake Musk, on-screen wallet addresses, a QR code, and a promise to double any crypto sent. More than 500,000 people watched, and hundreds of Coinbase customers tried to send funds within about two hours. No accounts were hacked; many were long-time users choosing to send money themselves. Coinbase's machine learning models score risky sends in real time, and once one is flagged, an agent talks with the customer, naming the scam pattern. The agent ran with no humans in the loop and had never seen this type of scam before. The FBI says Americans lost $11.4 billion to crypto scams in 2025. (Read More)
🫖 TEA For Thought: "Interesting that they deepfaked Musk to do a giveaway. MrBeast might be the first one who comes to mind for these types of activities."
🛠️ Skill of the Day
The Scam Pause: Gets a second opinion on a message, video, or offer before you send money or share a code.
Act as a calm fraud-prevention coach. Before I send money, share a code, or click a link, I want a second opinion.
What I received: [PASTE THE MESSAGE OR POST, OR DESCRIBE THE VIDEO OR CALL]
Who it claims to be from: [e.g. "Elon Musk on a YouTube livestream" or "my bank"]
What it asks me to do: [e.g. "send crypto to get double back" or "read back a code"]
Do the following, in order:
1. List every red flag you see, one line each, and quote the exact words behind it.
2. Tell me which known scam pattern this matches, if any, and how that scam usually works.
3. Give me one way to check it myself, using a contact method I already trust, not one from the message.
4. Rate the risk as low, medium, or high, and say what would change your rating.
5. Tell me exactly what to do next, and what not to do.
If you can't tell, say so. Never call something safe just because it looks professional or famous.Paste it into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own.
TEAHEE Moment
Stay sharp, stay informed. See you Monday.
If you enjoyed this TEA, follow along on social for more:






