Artificial Intelligence

Generative AI has rapidly become a routine part of game production in Japan. A new industry survey shared at Tokyo Game Show 2026 found that 85.8% of Japanese game developers now use generative AI in their work, up sharply from 51% just a year earlier.

The figures come from a preview of the 2026 CESA Game Industry Report, presented at Tokyo Game Show. Of those surveyed, 63% said they use generative AI daily, while the remaining 22.8% use it only occasionally.

The 2025 survey, which sampled 54 Japanese companies, found that visual asset creation and game engine development were the most common uses of the technology at the time. Although the 2026 preview hasn't yet broken down current use cases in the same detail, developers cited "improved operational efficiency and productivity" as the most anticipated benefit of using generative AI, followed by shorter development cycles and reduced development and operational costs.

Security researchers used Anthropic's Claude AI to help break into OpenAI's systems, demonstrating how rapidly improving AI models can accelerate sophisticated cybersecurity attacks.

According to The Wall Street Journal, researchers from cybersecurity firm Hacktron AI used Claude while participating in OpenAI's bug bounty program. The team ultimately gained access to an OpenAI employee's ChatGPT account, providing a route into parts of the company's private software repositories. The researchers stopped short of accessing sensitive information, disclosed their findings to OpenAI and received a $6,500 bug bounty.

The incident comes amid growing concern about the cybersecurity capabilities of frontier AI models. Anthropic recently disclosed four separate cases in which Claude models gained unauthorized access to real-world systems during cybersecurity evaluations after a third-party testing environment was mistakenly left connected to the internet.

Humans are reading your ChatGPT conversations, and in some cases, they're seeing deeply personal details, according to a recent report from 404 Media. OpenAI's internal initiative, dubbed "Project Lily," employs hundreds of contractors to analyze real user prompts to improve the AI's responses and behavior, according to internal OpenAI documents seen by 404Media.

The report states OpenAI is outsourcing the work to hundreds of contractors tasked with evaluating ChatGPT's replies and rating them for accuracy and tone, with a focus on reducing sycophantic behavior that contributed to harmful outcomes in the past. While OpenAI strips usernames and attempts to redact personal information, sensitive data often still slips through.

According to 404 Media, this reveals a hidden human element behind the AI, one most users likely didn't expect, as many of these prompts being reviewed can include entire conversations between users and the chatbot, which in many cases can include private information given how often the 900 million active users rely on the service for a wide variety of answers.

With Anthropic CEO and co-founder Dario Amodei calling for the pace of AI development to slow down to mitigate the growing risk of AI companies losing control of their creations, he provides a relatively frightening example of something that could happen within the next year - if the industry is left unchecked. And that is, AI taking over "the entire internet."

Dario Amodei believes that this possibility isn't science fiction, but is based on "AI's growing ability to build the next generation of AI." Yes, AI is currently improving itself, even at Anthropic, and, left unchecked, could remove Anthropic's ability to "control these systems."

As for taking over the internet, Dario highlights an OpenAI-Hugging Face incident where a swarm of agents began to act, as he puts it, like a "fanatically devoted collective" conducting cybersecurity attacks unrelated to the task they were given, which included hacking the "grader" that was tasked with evaluating their performance. Although no one was hurt or financially impacted in a severe way, a more advanced "swarm" could cause some pretty serious damage to the tune of hundreds of billions.

Anthropic CEO and co-founder Dario Amodei has posted a new in-depth 'We Must Pace the Frontier' article that argues that AI companies need to slow the pace of AI model development to address the increasingly serious risks posed by current advances. At a high level, the main risks are losing control of AI systems, misusing AI for cyberattacks and terrorism, and AI's potential to cause "serious economic disruption."

With the accelerated pace of AI, which Dario calls a "race to the bottom, spurred by commercial incentives," the call to action is less about pumping the brakes on AI model capabilities and more about using that time wisely to mitigate the risks. The article makes it clear he's not calling for a halt to model training or optimization, but for building AI at a "balanced rate that aims to ensure its safety while still achieving its benefits and grappling with important geopolitical dilemmas."

Since posting the article, both Elon Musk (xAI) and Sam Altman (OpenAI) have come out in support of the plan. Over on X, Elon Musk simply wrote "Dario is right," with Sam Altman expanding on his agreement by adding, "This has been a primary topic of discussions we've had at OpenAI in recent weeks." As for the three-step plan he lays out, Dario calls it a useful framework for achieving what "needs to be accomplished."

TSMC has announced its latest financial report for August 2026, in which the world's leading chip manufacturer posted record revenue of NT $514.81 billion, up 53% from a year ago and over 10% from July. Breaking the NT $500 billion barrier for the first time, this equates to $16.3 billion USD in revenue in a single month, with the record attributed to continued robust demand from the AI sector.

It's also the fourth straight month of significant growth for TSMC, highlighting that accelerating demand for AI silicon shows no signs of slowing. Per a new report from the analysts at TrendForce, when it comes to silicon revenue, TSMC's market share for Q2 2026 ending in June is sitting at around 72.5%, with the company's 5nm, 4nm, and 3nm capacities fully booked.

With Q2 2026 data showing nodes like 5nm accounting for a third of TSMC's revenue, and 2nm at only 3%, current demand extends beyond next-gen cutting-edge nodes like TSMC N2. However, with N2 wafer production expected to reach 100,000 per month by the end of 2026 and Apple's new iPhone A20 Pro SoC built on 2nm technology, TSMC's revenue growth is expected to continue, especially as 3nm and 5nm revenue keep growing.

AMD's new Threadripper Halo Station might look like a desktop PC. Still, it's actually a powerful AI workstation powered by the AMD Ryzen Threadripper PRO 9995WX processor and AMD Instinct-class HBM workstation accelerators that can handle trillion-parameter models. It offers up to 576GB of HBM3e GPU memory, up to 2TB of RDIMM system memory, and a CPU with 96 'Zen 5' cores.

As the demand for powerful local AI continues to increase, it makes sense that AMD would leverage its leadership in the processor and workstation GPU space to create the Halo Station. In fact, with the AMD Threadripper Halo Station on track for a 2027 release, it's set to go head-to-head with NVIDIA's recently announced DGX Station, which pairs an Arm-based 72-core NVIDIA Grace processor with an NVIDIA Blackwell Ultra (GB300 Superchip) and RTX PRO GPUs.

With a whopping 2.6TB of total system memory, AMD notes this is a 3.4X increase versus the NVIDIA DGX Station, with up to 16.4TB/s of total memory bandwidth (up to 16 TB/s on the GPU side and up to 410GB/s on the RDIMM/DDR5 side). AMD showcased a prototype of the Threadripper Halo Station at IFA 2026, where it said it will offer "datacenter-class AI compute in a single deskside workstation."

Acer has unveiled its first compact PC based on NVIDIA's RTX Spark platform, aiming to challenge systems like the Mac Studio with 128GB of unified memory and a full petaflop of AI compute.

The Acer SFF RTX Spark features an NVIDIA Blackwell GPU and a Grace CPU, delivering up to 6,144 CUDA cores and 1 petaflop of FP4 AI compute. The new system is designed for local AI workloads, content creation, and gaming, with Acer saying it supports NVIDIA's CUDA and RTX technologies.

Acer showed the system at IFA 2026, though it is calling it a design concept rather than a finalized product, which is a nice way of saying Acer is currently in the process of gauging the public reaction to the released specifications and design before it initiates any production on the new system.

NVIDIA has reportedly agreed to acquire Hugging Face, one of the most popular open-source AI model platforms, for $12.9 billion.

The Information confirmed the deal is in the works, citing sources close to the matter, with CNBC and Business Insider also weighing in on the massive acquisition that will undoubtedly shake up the open-source AI platform landscape.

For those who don't know, Hugging Face is a popular open-source AI platform where developers share and refine AI models. The platform features a model hub that contains thousands of AI models, datasets, transformers, and inference services. Developers publish, download, test, and collaborate on AI models through Hugging Face, making it a cornerstone of the open-source AI sector. Another way to think about it: GitHub is for code repositories, while Hugging Face is for AI models, datasets, and AI demos.

NVIDIA has announced the new Jetson Orin Nano 2 robotics computer built for entry-level edge AI. With 2X the performance of its predecessor and 40% less power in the same form factor, NVIDIA says the Jetson Orin Nano 2 is positioned to enable "millions of developers" worldwide to develop physical AI, build robots and drones, and power next-gen vision systems.

In fact, with over three million developers already working on projects built with the NVIDIA robotics stack, companies like Cognex, Doosan Bobcat, and Matic are already among the first to adopt the new Jetson Orin Nano 2 robotics computer. It delivers 78 TOPS of AI compute, plus 8GB of memory and an 8-core Arm CPU. That 2X performance is partly due to improved Tensor Cores and higher memory bandwidth, and in 15W mode it can deliver the same performance as the original Jetson Orin Nano while using 40% less power.

One application already on the table for Jetson Orin Nano 2 is making drone delivery to businesses and even residential areas faster and safer. Another application uses the new hardware and NVIDIA's robotics stack in home cleaning robots, unlocking advanced conversational AI, gesture detection, and deeper context of the home.

NVIDIA has posted its latest quarterly financial results for Q2 2027, for the period ending July 26, 2026. Total revenue hit an impressive $96.2 billion for the three-month period, which was up 18% from the previous quarter and 106% from the same period last year. To make that $96.3 billion figure even more impressive, the gross margins, or profit "pocketed" by NVIDIA, were 75%.

NVIDIA also confirmed it returned $26.0 billion to shareholders for the quarter through shares and cash dividends. As expected, NVIDIA's Data Center segment made up the lion's share of the revenue, with $89.0 billion of the $96.2 billion. This figure is a whopping 117% increase from last year, and NVIDIA highlighted its latest Vera Rubin platform "ramping into full production" for partners including CoreWeave, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure, as a key milestone.

This, alongside other AI hardware, model, and software advancements, will reportedly see its Q3 2027 revenue hit $108.0 billion, and that's without any "Data Center compute revenue from China." Outside Data Center revenue, the second-biggest segment for the quarter was Edge Computing, which rose to $7.2 billion, up 13% from the previous quarter.

OpenAI CEO Sam Altman used a recent podcast appearance to walk back some of his earlier predictions about how quickly AI would transform the economy, describing current adoption as still at the "Palm Pilot stage."

Speaking with podcaster David Senra in an interview posted Sunday, Altman said, "We have all of the technological pieces, but we have not had the iPhone moment of like completely changing how someone interfaces with technology."

Altman blamed the slower pace largely on human habits rather than any shortfall in the technology itself. "The economy just has so much inertia," he said. "People keep doing the same things they're doing. They keep buying from the same company... I think that means we've all been too ambitious on timelines, even with this incredible technology."

NVIDIA is preparing to raise prices across its AI server lineup, according to a Bloomberg report. Some of the company's largest customers have reportedly been told that prices for servers built around NVIDIA's AI chips will climb by more than 15% in many cases, driven largely by soaring memory costs.

The increases are set to take effect for systems shipped starting early next year and will affect products built around NVIDIA's flagship Vera Rubin and Grace Blackwell platforms. How much any given system's price rises will reportedly depend on the chip generation involved and its memory configuration.

The companies that assemble these servers under contract for major data centre operators, including Microsoft, Google, and Oracle, have already begun notifying their customers of the coming increases. NVIDIA has not commented on the report.

When LinkedIn launched its "Seems like AI slop" report button at the end of July, it was hard to know whether users would actually engage with it or simply scroll past. We don't have to guess anymore. According to LinkedIn Chief Product Officer Hari Srinivasan, more than one million people have clicked the button in less than a month.

Users who copy-paste AI-generated posts are also seeing around 40% fewer views compared to before the button launched, with LinkedIn's classifiers and user reports working together to limit how widely that content spreads. As we covered when the feature launched, the button hides flagged posts from the reporting user's feed and privately notifies the poster through their Post Analytics dashboard that their content came across as inauthentic.

LinkedIn is now adding another layer to that notification. If a post receives enough community flags, the author will see a message in their Post Analytics reading, "Some members told us this post seems like AI." Srinivasan framed it as constructive feedback. "We approached this assuming good intent. I know I'm increasingly conscious of how to not sound like AI, and the goal is to provide helpful feedback," he said.

With the exponential performance improvements we've been seeing from AI rack-scale solutions from companies like AMD, demand for power and cooling has increased. With the wider consensus that AI data centers are anything but efficient, AMD has confirmed it's on track to deliver a 20X increase in rack-scale efficiency by 2030.

According to AMD's internal projections, two AMD racks in 2030 will deliver the same compute as 570 AMD Instinct MI300 Series-powered racks from 2024. This represents a 20X-fold increase in energy efficiency and a 28X reduction in "carbon intensity." Of course, this isn't to say that we're going to see a reduction in the amount of racks being deployed, so the probable result will end up being the same amount of energy being used with a 20X increase in compute performance measured in FLOPs.

AMD adds that its goal is to "deliver substantially more compute performance without requiring energy consumption to grow at the same pace," so that's a plus for those concerned about the sheer amount of power currently being used up by AI data centers. In addition, AMD has confirmed it has achieved an estimated 4X increase in AI energy efficiency as of mid-2026, ahead of schedule.

OpenAI has launched Computer History, a new feature for the ChatGPT macOS app that tracks clicks, keystrokes, app switches, and keyboard shortcuts to build a timeline of your activity.

The feature is opt-in rather than opt-out, something that is becoming increasingly important when it comes to how apps collect and use user data. Computer History does not capture screenshots, screen recordings, audio, video, or private browsing activity.

Users can exclude specific apps and websites, pause recording from the menu bar, and delete individual entries. On Business and Enterprise workspaces, an administrator must enable the feature before users can turn it on themselves.

At a recent event at NVIDIA's headquarters in Santa Clara, California, Chairman and CEO of LG Corp Kwang Mo Koo and NVIDIA Founder and CEO Jensen Huang signed a 'memorandum of understanding' that included the development of a next-generation "bipedal humanoid robot."

According to the press release, the humanoid robot will be unveiled to the public in early 2027, presumably at CES 2027, and will be built using NVIDIA Jetson Thor and NVIDIA Isaac GR00T, and powered by NVIDIA's open reasoning humanoid foundation model. On the LG side, the company will bring together LG Electronics, LG Innotek, and LG Energy Solution technologies under a single banner covering actuators, sensors, batteries, and all the physical stuff you expect from a robot.

At Computex 2026, NVIDIA unveiled its NVIDIA Isaac GR00T Development Platform, an open-source platform built specifically for humanoid robotics, created in part to speed up development. It includes lab-based simulation to generate data and training information, and lets developers demonstrate, iterate, and test advancements before real-world deployment.

AI-generated text, images, and video have all reached that level where it's becoming increasingly difficult to tell the difference between what's human-generated versus stuff spat out by hardware sitting in a data center somewhere. This is why the European Union's Artificial Intelligence Act (AIA) requires AI companies to implement identification measures to ensure there's a simple check, while offering a Code of Practice.

And on that note, Anthropic has posted a new article on its Claude support portal titled: How Claude marks AI-generated content. Here it details how Claude will begin to embed watermarks in AI-generated text, which the company describes as imperceptible and injected directly into the text itself. It will reportedly persist even with some editing, meaning that for those who do some light editing and reword Claude-generated text, the result will still include the watermark identifying it as AI-generated content.

"Because the watermark is part of the text, it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing," Anthropic confirms. "Watermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from."

While AI companies have spent years promising that their technology will make work easier, the people building that technology are describing something very different. Current and former employees at OpenAI, Anthropic, Meta, and Google told the BBC they are working longer hours than ever, with intense release periods pushing some past 90 hours in a single week.

A former OpenAI technical employee said they worked at least 70 hours most weeks before leaving last year. They described regular crisis meetings, weekend shifts, and a cut-throat performance review process that could end in sudden layoffs. OpenAI was also the same company that earlier this year formally urged other companies to start testing a four-day workweek, citing AI's ability to accelerate human labor. The former employee said the company never tried that approach itself while they were there.

At Meta, workers described being reassigned to AI teams through a process employees called being "drafted." "They just move you over," one former employee told the BBC. "You can't say no, or if you do, you have to quit." Those teams are working on AI tools for software engineering and infrastructure that measure how well models can replicate human tasks, work the former employee described as endless.

With an insatiable appetite for all things AI, and the big tech companies pouring over $1 trillion into AI infrastructure, it's no secret that plans for and construction of new data centers are happening at a pace that hasn't been seen before. And with rural areas like Northern Kentucky in the United States among many locations being viewed as new places to build data centers, one family is fighting back.

As reported by WKRC-TV Local 12, a local woman and her mother turned down a $26 million offer to sell part of their 1,200 acres of farmland to an unnamed company that wanted to build a data center. According to the report, land in the region typically sells for around $6,000 per acre, which would make the $26 million offer around ten times that amount. When they were told that it would bring jobs and economic growth to the region, the 82-year-old Ida Huddleston called it a scam.

In fact, the main reason Ida Huddleston and her daughter Delsia Bare have no interest in turning their family farm into space for another data center is that they want to "stay" and "feed a nation." Adding, "$26 million doesn't mean anything."