Silicon Valley's next big AI idea is probably already sitting in a 1991 paper by Sakana AI's new chief advisor. The Tokyo-based startup has hired Jürgen Schmidhuber, whom it calls the "father of modern AI," as Chief Scientific Advisor. Sakana credits him with early work on world models in 1990 and with the 1991 deep learning techniques behind much of today's AI boom. His 1987 thesis on meta-learning and recursive self-improvement, where AI keeps improving itself, started nearly four decades of research into how machines can learn to learn. Sakana says those ideas directly shaped its projects like the Darwin Gödel Machine and The AI Scientist.
Japan is the birthplace of foundational neural network architectures and advanced robotics. It is a privilege to join Sakana AI to help bridge these two worlds. The future of intelligence is not just language; it is physical AI powered by world models. Sakana AI is uniquely positioned to reclaim Japan’s legacy of innovation by building autonomous systems that can truly understand, simulate, and interact with the physical universe.
Jürgen Schmidhuber
Schmidhuber will help steer Sakana's new RSI lab, which aims to build a self-reinforcing research loop that makes machines smarter. To do that, Sakana wants to gather a "critical mass of world-class experts" in Tokyo, it says.
Google wants to put AI data centers in space. Its "Suncatcher" project would run AI infrastructure in orbit on solar power. On October 1, an experimental satellite called "MVP" is set to launch on a SpaceX Falcon 9 from Vandenberg Space Force Base in California, the New York Times reports. The "refrigerator-size satellite" was built and tested at Planet Labs in San Francisco. Google VP James Manyika says it should pack enough computing power to handle basic AI queries from orbit.
SpaceX and Blue Origin are also working on orbital data centers, but the hurdles are massive. Google manager Travis Beals estimates you'd need around 10,000 satellites to match a single 1-gigawatt data center on Earth. There's no air in space for cooling, cosmic radiation can fry chips, and launch costs would have to drop to about $200 per kilogram before the math works out. Jeff Bezos thinks it could take up to 20 years before orbital data centers beat ground-based ones on cost.
Black Forest Labs is releasing FLUX 3 Action, an open AI model designed for robotics. It builds on the multimodal FLUX 3 model, which was trained mainly on video but also on image and audio data. As a world-action model, it takes multi-camera video feeds from a robot workspace and predicts what action an agent should take next and how the environment will change as a result.
On the RoboLab-120 leaderboard, FLUX 3 Action sets a record for success rate with just seven billion parameters, according to BFL, less than half the size of the previous best open model while running up to 3.95 times faster. Large reasoning models can plan well, BFL says, but they're often too slow and too bulky for robots, making this kind of efficiency critical for on-device deployment. BFL also sees potential in digital environments, where video games serve as testing grounds for navigation and fast-reacting agents that operate computers could follow down the line. The weights are available on Hugging Face.
Anthropic's AI model Claude found a previously unknown enzyme system in DNA databases, doing most of the analysis on its own. Humans set the research question and ran the lab experiments. The system resembles the gene-editing tool CRISPR.
Anthropic calls it one of the first discoveries from its Bay Area biology lab, which opened in spring 2026. About 950 AI agents spent 21 hours combing through more than 200,000 reverse transcriptases, enzymes that convert RNA into DNA. Next to one unusual enzyme, they found repeating DNA sequences. The system, dubbed "ART," shows up mostly in bacteriophages, viruses that infect bacteria. Its function is still unknown.
Lucas Harrington is less impressed. He earned his PhD under Nobel laureate Jennifer Doudna, who co-discovered CRISPR-Cas9, and co-founded Mammoth Biosciences with her. Mammoth uses genome mining to hunt for CRISPR systems itself. On X, Harrington wrote that the method has been around for decades and that similar systems have been known since 2008. The hard part is figuring out what a system actually does, and Anthropic hasn't shown that. Presenting early results as a major discovery isn't helpful, he said.
AI is hitting a fixed benchmark performance level at a rapidly falling cost. Epoch AI measures a price decline of about 13x per year. After stripping out hardware gains and competition, MIT puts annual algorithmic progress at about 3x. That doesn’t mean today’s best models are cheaper, though. Reasoning models, for example, can cost more because they use far more compute per task. When picking a model for real-world use, quality, speed, and error rate matter just as much as price.
Senator Bernie Sanders and Representative Greg Casar introduced a bill on September 23 that would permanently ban the development and use of artificial superintelligence. The Ban Artificial Superintelligence Act would also immediately freeze development of advanced AI systems until a new federal agency can put clear safety rules in place.
Current AI models can already hack computer systems, create new types of viruses, and build new AI on their own, while companies pour hundreds of billions of dollars into the technology despite known risks, Sanders says. AI poses an existential threat to humanity, and a handful of tech billionaires shouldn't get to write their own rules. Casar warns that AI superintelligence could kill countless people, while Donald Trump wants to speed up its development.
The bill also calls for a new cabinet-level AI agency. Companies that violate the law could face forced dissolution. Individuals could face up to 20 years in prison, a penalty on par with illegal nuclear weapons development. On top of that, the legislation pushes for international agreements to prevent superintelligence development worldwide. Both the U.S. and Chinese experts have recently floated similar AI regulation efforts.
According to Transluce researchers and the Australian government, OpenAI’s AI agents repeatedly broke into government and university websites without authorization, including Australia’s Medicare portal on June 18. The cause was a mundane data search. Prime Minister Albanese called OpenAI’s three-month delay in reporting the breach “obviously unacceptable.” Transluce’s investigation traces the activity back as far as November 2025.
Meta expanded its AI agent Muse and showed off several new devices at Meta Connect 2026.Muse now supports real-time video chats with customizable avatars, gets its own email addresses, and can control Mac apps. In the coming months, Muse will also run on Meta's smart glasses. There, it will turn conversations into actions, like adding items to a shopping list or scheduling meetings.
Mark Zuckerberg also unveiled the Muse Charm, a Tamagotchi-like device with a 2-inch display, 5G, and its own operating system. It's expected to ship by December. CTO Andrew Bosworth called the device an exploration and said it isn't the core of Meta's AI hardware strategy.
Meta also announced new Ray-Ban Meta Audio Glasses, which drop the camera and start at 349 euros. The third-generation Ray-Ban Meta starts at 449 euros, and a pair of VR glasses will cost $1,299 when they arrive in spring 2027. On privacy, Meta plans to bring its "Private Processing" system to the glasses. The system is designed to shield user data even from Meta itself, but the company hasn't given a timeline.
