Osaka University, the National Institute of Information and Communications Technology (NICT), and Hamamatsu Photonics announced on September 1, 2026, a "10-fold multiplexed quantum optical interface" that simultaneously detects photons from 10 neutral atoms arranged a few micrometers apart using a 32-channel integrated optical waveguide array. The multiplexing degree of 10 is a world record, and in principle it could be extended to about 100-fold multiplexing. The paper was published in the optics journal Optica on August 31.
Space One announced on August 30, 2026, that the cause of the flight termination of the small rocket "Kairos" No. 3, which occurred 68.8 seconds after its March 5 launch, was electrical noise from static electricity that had accumulated on the vehicle. The noise rebooted the communication software, temporarily cutting off communication, which caused the autonomous flight safety system to terminate the flight.
By combining two neural networks, researchers shortened semiconductor internal analysis that used to take several days to under one millisecond. The training data was about 400 cases, roughly one-hundredth of the conventional amount.
A team at UC San Diego created about 500,000 variants of a DNA sequence called an initiator, measured them all at once, and used machine learning to decode the patterns. It was predicted that about 60% of human genes have an initiator.
The National Institute of Informatics released a Japanese large language model with about 33.2 billion parameters, under the Apache License 2.0, along with the DPO dataset used for training.
The National Institute of Informatics announced on August 18, 2026, that it has integrated approximately 4 million research outputs from Japanese institutional repositories into OpenAlex, one of the world's largest academic databases.
A survey of 5,083 university students found that those with stronger tendencies toward internet dependence also had a higher risk of gambling problems. The research team proposed that 'digital literacy education should start from elementary and junior high school.'
An analysis method was developed to decipher what clues an AI, trained on 2,681 types of inorganic materials, relies on to predict optical properties. The AI was found to group similar materials on its own and identify relationships between important elements or atomic arrangement and the properties.
The Ministry of Education proposes expanding information education; classes on the risks and use of generative AI are expected to begin phasing in from fiscal year 2028.
A molecular structure that behaves like a liquid droplet has been developed using light-responsive DNA. As a "light-driven molecular device," it is expected to find applications in medicine and nanotechnology.
Combining brainwaves with AI lets people operate machines just by thinking. While transforming medicine, this has also sparked discussion of neurorights aimed at protecting 'freedom of thought.'
A class where students assemble commands in sequence to fly a drone. Using teaching materials with visible results, children learn trial and error and logical thinking.
A probe needle as thin as a human hair, at 0.08mm, has been developed. Japanese precision shines through in the testing technology that underpins the quality of cutting-edge semiconductors.
A data center in Ishikari, Hokkaido, is powering itself with renewable energy. Facing surging electricity demand driven by generative AI, its location strategy combines outside air cooling enabled by the cool climate with Hokkaido's abundant renewable energy resources. PUE (Power Usage Effectiveness) is the key metric.
[What Happened] South Korea's research institute ETRI announced a plan to invest about 47 billion won over four years to lead the next-generation '6G' communications field in AI-native design (designing networks with AI built in from the start). Countries and companies worldwide are intensifying competition over international 6G standards.
[Why It Matters] Communication standards change generation roughly every ten years, with 6G following 5G. 'AI-native' refers to a design philosophy in which the network itself is autonomously optimized and controlled by AI, rather than having AI added on afterward. Countries and companies that control the standard gain major advantages in patents and industry, making this both a technological and a national strategic competition.
[Going Deeper] For 6G, the use of extremely high-frequency radio waves in the terahertz band is under consideration. Higher frequencies can carry more information, but they travel in straighter lines and are more easily blocked by obstacles—a physical trade-off that research using AI-based radio wave control and large numbers of antennas aims to overcome.
[Think About It] In an era when networks think and optimize on their own, what decisions or rights do you think we as users must not give up?
A research team at the University of Minnesota has discovered a phenomenon in which changing the thickness of a metal thin film by just a few nanometers dramatically alters its electronic behavior. This finding enables atomic-level precision control in the design of semiconductor devices and can be applied to materials design for next-generation processors and quantum computers. It could provide a clue for technological innovation that goes beyond the limits of Moore's Law.
What sets AI agents apart from previous AI is that they 'think and act.' For example, if you entrust stock trading to one, it can trade faster and in greater volume than a human—but if it runs out of control, losses can occur in an instant. In medicine too, while it helps with diagnosis, the question remains of who makes the final decision. What matters here is a mechanism where a human makes the final check ('human-in-the-loop'). This connects to the idea that SciCirc values: not letting AI complete a process entirely on its own, but always involving a human. More than technological progress itself, 'creating rules for how it is used' is the core issue for the information society going forward.
When you use AI, the computation is actually taking place 'somewhere in the cloud (a massive data center).' This announcement is similar to a story where 'more power companies appeared, making electricity easier to buy.' When the same AI can be used across multiple clouds, you can choose based on price and security, reducing dependence (lock-in) on any one company. 'Cloud computing,' which you learn about in information technology class, is exactly the main battleground of today's industrial competition. Reading the news from the perspective of 'who controls the pipes that AI runs through' changes how it looks.
There are two advantages to running AI on a local device rather than the cloud. (1) Privacy: since data is not sent externally, the risk of information leaks is reduced. (2) Speed and connectivity: it works without delay even without an internet connection. At the same time, having powerful AI in individuals' hands also raises the risk that it could be misused more easily, for example to generate disinformation. This is a theme for considering the essence of information itself — how the safety of society can change simply based on where computation takes place.
Until now, software has been built on the assumption that "a human presses the button." In the age of agents, operating systems are being redesigned on the assumption that "AI presses the button on the human's behalf." This marks a major turning point in the history of automation. At the same time, allowing AI to freely touch files and applications means that security and permission management (who authorized what) become more important than ever. It is a useful topic for thinking about how convenience and risk go hand in hand.
We have entered an era in which the value of AI companies is determined by expectations of "how much work they will be able to take on in the future." The figure of $965 billion is evidence that the market views AI as becoming a piece of social infrastructure on par with search engines and the cloud. Once listed, financial information will be made public, allowing us to see the reality of AI businesses (revenue, costs, and the scale of losses) for the first time. It is an ideal topic for practicing how to read technology news through the lens of "the flow of money."
Under the EU's AI Act, full-fledged compliance obligations for 'high-risk AI' used in fields such as healthcare and hiring will take effect in stages starting in August 2026. Companies that violate the rules could face steep fines of up to 7% of their global annual revenue or €35 million, whichever is higher.
A survey by the professional association ISACA found that deepfakes and cyberattacks made more sophisticated by generative AI have emerged as the top concern for 2026. The share of respondents citing this as a threat reached 59% worldwide and 60% in the Asia-Pacific region, underscoring the growing importance of information literacy in distinguishing the real from the fake.
Japan and the EU held a ministerial-level meeting in Brussels and announced a new strategy for cooperation on AI, semiconductors, and quantum technology, as well as on building information infrastructure such as submarine cables. The aim is to build a resilient supply chain that does not depend on any specific country.
Chrome 149, released by Google, is a major update that fixes a total of 429 vulnerabilities, including 22 rated at the highest severity level, Critical. It includes serious bugs such as out-of-bounds read/write in ANGLE and use-after-free (UAF), which could allow arbitrary code execution if exploited. This provides material for understanding browsers from the perspective of the software stack.
The World Semiconductor Trade Statistics (WSTS) organization released its spring forecast, projecting that the global market in 2026 will reach $1.5112 trillion, an increase of approximately 90% year-on-year. AI-related memory ICs are the main driver, with a projected 3.5-fold increase. The Japanese market is also forecast to grow 27.6% to $57.1 billion, as AI and data center investment pushes the industry higher.
GitHub has overhauled Copilot's billing structure, moving from a request-count basis to a token-consumption-based system called 'GitHub AI Credits.' While the base fee remains unchanged, using AI agent mode or high-performance models now consumes additional credits, making cost management an important issue for AI-assisted coding.
Rakuten Securities announced that it will phase in mandatory passkey (FIDO2-compliant) authentication for login starting in June 2026. Biometric information is stored on the device and never sent to the server; as unauthorized access to securities accounts has become a social problem, the shift to passwordless authentication is accelerating across the financial industry.
IBM and Red Hat announced 'Project Lightwell,' a $5 billion initiative that will use more than 20,000 engineers and AI to detect and fix open-source vulnerabilities at scale. Financial institutions are participating as early adopters. The project is drawing attention as a new model for addressing the risk that open-source vulnerabilities can spread throughout the entire supply chain.
The technology behind the news
Brute-Forcing About 500,000 Variants Through Machine Learning Revealed 'Start Signals' in About 60% of Human Genes