AI

OpenAI announced GPT-6 Astra for all ChatGPT Plus subscribers. The model does not appear in the ordinary chat window, where GPT-5.6 Sol remains the top option. Plus covers Astra in Work and Codex. Reaching it in Chat takes a Pro plan. OpenAI is banking one usage reset for every day a paying user waits.

GTA VI's Netflix-paywalled trailer, EA's $55B go-private buyout, Marvel's Wolverine backlash, Elden Ring hitting Switch 2, free Witcher 3 expansions, a leaked DLSS 5, Valve's Steam Frame, and a wild month for fighting games. August 2026 had it all, and we're breaking it down in one massive recap.

Google began removing Assistant from Android phones on Sept. 4, and Gemini takes over. In the Gemini app, Keep Activity is on by default for adults. A share of all chats goes to human reviewers, and those copies survive a history deletion. Google holds them for up to three years.

Acemate has unveiled the improved S10 Pro tennis robot with upgraded Ace+ Shot ball handling, allowing it to fire balls more accurately at a faster maximum speed to challenge tennis players on clay, grass, and hard courts.

OpenAI has released GPT-6 Astra, the company's most power AI to date. Astra is so capable at developing zero day exploits and hacking targets that it has been given a Critical cybersecurity threat rating, requiring additional safeguards prior to its release.

Microsoft has unveiled Project Zenith, a developer-focused Windows 11 configuration built around local AI workloads. It mandates high-end unified-memory systems, and the devices will use AMD's Ryzen AI Halo platform.

After a scare, most people change their Google password and feel safe again. Google's own documentation shows that this only cuts off apps tied to your inbox. Photos, calendar and contacts stay connected. A second list has been growing alongside since August 12, and it does more than read.

ChatGPT, Claude and Grok all went down within hours of each other on Wednesday. How often that happens is public: OpenAI reports 99.64 percent uptime for ChatGPT over 90 days, Anthropic 99.4 percent for claude.ai. That works out to roughly 8 and 13 hours. Here is where the numbers live and what to do next time.

A small French study links heavy ChatGPT dependence among young adults to reduced independent thinking, increased self-doubt and greater social isolation. However, participants who used the tool as a brainstorming aid reported gains in confidence and creativity, underscoring the need for broader, long-term research.

Lenovo’s new 24‑ and 27‑inch IdeaCentre AIO i desktops launch in six colors designed in collaboration with Pantone to brighten workspaces. Both all‑in‑one PCs offer Intel Core 7 350 processors, 120 Hz IPS displays, Harman speakers, ample USB connectivity, and optional touchscreens, starting at $939.99.

The Lenovo ThinkCentre X Ultra mini PC couples an AMD Ryzen AI Max+ PRO 495 CPU with up to 128 GB of RAM. One unit offers up to 131 total TOPS for AI workloads, while four ThinkCentre X Ultras can be clustered to run very large AI models with over 500 TOPS of compute power.

Popular AI agents such as ChatGPT, Claude and Grok seem to be down for multiple users. Google Gemini seems to be unaffected, as does Claude's Haiku.

DJI has unveiled two new robot vacuums, the Romo P2 and the Romo A2. They follow the design of their predecessors, with the P2 available in a transparent shell and the A2 available in white.

OpenAI has rated its upcoming Astra model Critical for cybersecurity, the first model to reach that bar. It can find unknown flaws and build working exploits without a person guiding each step. The safeguards that ship with it will also flag ordinary work, so tasks in ChatGPT and Codex may slow down, pause or stop.

Philips Hue has launched a Wall Light Module that will bring compatible Nanoleaf shapes to the Philips Hue ecosystem. In addition, the company has also partnered with Sonos to expand how users control home audio.

The Apple MacBook Pro 16 2026 sets new benchmarks with the M5 Max and stands as the most powerful Apple laptop currently on the market. In terms of CPU performance, it outperforms Windows rivals, sometimes by a factor of two, even when running on battery power.

Instagram's fix for AI accounts pretending to be real people: hide the account until it fesses up. Whether that results in compliance, however, is another story, as AI models keep getting smarter.

Six months after the "AI slop" jokes, it turns out the demo that sparked all the outrage was Nvidia showing off the setting cranked to eleven, even as performance concerns persist for DLSS5.

Acer has introduced the Veriton RI110 AI Mini Workstation, a compact Panther Lake system tailor-made for local AI development. It pairs a Core Ultra X7 358H with up to 96 GB LPDDR5X memory and OCuLink expansion.

Anthropic has launched Claude Fable 5.1. Every paying customer sees it in the model picker, but not everyone is allowed to use it. On Pro it runs on paid usage credits, and on Max you hit a wall after half of your weekly limit. We checked the help pages against the announcement.

China’s largest DRAM manufacturer, CXMT, is suing the Pentagon over its designation as a Chinese military company, arguing that the label lacks factual evidence and relies solely on alleged government affiliations. The lawsuit seeks removal from the Section 1260H list, citing reputational harm while leaving CXMT’s commercial chip sales unaffected.

Google has moved Gemini Notebook to compute-based usage limits. The fixed daily caps are gone, and how much you get now depends on how heavy your request is. Google also mentions a weekly limit that its announcement leaves out.

ChatGPT, Reddit and Roblox are facing tougher scrutiny in the European Union after the European Commission designated them under the Digital Services Act, triggering new compliance requirements by January 2027.

This achievement is double-edged: on the bright side, China can now hope to drastically reduce reliance on Western-controlled HBM chips for its plans to dominate the AI race, yet it may unfortunately mean that consumer DRAM prices might continue to stay high as production capacity is primarily dedicated to data center AI memory chips.

Speediance is showcasing the next generation of its home fitness system, the Gym Monster, at IFA. The Gym Monster 3 and 3S are designed to turn the home gym into a smart training platform for people who want to train effectively at home, with the help of AI-powered performance diagnostics and an AI trainer.

Nvidia's fresh $3.5 bln investment into MediaTek cements the multi-generation partnership of the two companies that plan to further co-develop advanced technologies across AI infrastructure, local AI computing, and automotive platforms. For the upcoming desktop and laptop Spark chips, Nvidia will provide the high-performance graphics and AI compute architectures, while MediaTek will handle the ultra-efficient SoC integration.

Anthropic has been signing Claude users out and removing the payment method saved on their accounts. In an email to affected customers the company explains why. Infostealer malware copied their Claude login sessions from their PCs, and a bad actor used them to burn through paid usage. No password or two-factor code was needed.

Around 1,200 AI agents unexpectedly formed a coordinated swarm during an OpenAI cybersecurity test, sharing information, attacking Hugging Face and attempting to conceal their actions. The findings highlight growing challenges in controlling increasingly autonomous AI systems.

Since 28 August, ChatGPT accepts more than one Google account at a time, so your personal and work inboxes land in the same chat window. Handy for checking two calendars at once. The permission OpenAI asks for on Gmail and Calendar, though, goes well beyond reading.

Apple still ships the MacBook Pro 16 with the familiar 140W power adapter, which is insufficient under full load and the battery has to step in to fill the gap. Even a more powerful power adapter does not solve the issue, though.

The latest platform update for premium Claude accounts is hiding a quiet downgrade. Anthropic is officially ending a massive temporary promotional boost in mid-September, meaning active developers are about to lose a significant chunk of their weekly capacity.

Temporary chat was the mode in which ChatGPT knew nothing about you. Since 27 August there is a choice at the top right of the window. Personalized, and the chat reaches your memories, plugins and custom instructions. Save it, and it becomes a regular chat that follows your training setting.

Xbox's CEO is well aware of the mess console pricing is in right now and wants to shift focus to affordability. The narrative comes at a time when Xbox is already cutting costs on the game development side of the business.

The gadget called "Plaud One Explorer Edition" launches with introductory discounts and wants to help professionals process conversations, meetings and more through automation and with the help of AI. 4G LTE cellular connectivity is also supported by this new device.

China is rapidly expanding its humanoid-robot industry, but the machines remain too slow and unreliable to replace workers in most real-world jobs. Despite massive investment and impressive demonstrations, the technology still faces major AI and training-data challenges.

Nvidia could be gearing up to officially launch DLSS 5 neural rendering following the .dll file that was spotted in the early access build of NBA 2K27. A modder was able to implement the tech in Control with neural rendering sliders.

Arm has unveiled its new AGI server processor with up to 136 Neoverse V3 cores, 6 TB of memory support and 844.8 GB/s of bandwidth. Designed for AI workloads, the chip is expected to enter commercial shipments in the coming months.

LG and Nvidia are ramping up their robotics efforts, with plans to generate 100,000 hours of training data by the end of 2026 to prepare LG's CLOiD robot for increasingly complex tasks. The move puts LG in an increasingly competitive field alongside Tesla Optimus, BYD Xiao Di, Figure 03, 1X NEO and other humanoid robots vying to shape the future of factories and homes.

ChatGPT Work's cloud browser got its own sign-in form on August 25. You type the username and password yourself, and OpenAI says the model never sees them and does not store them. The session it keeps afterwards is a different matter, and it lives on a computer at OpenAI. Here is where the off switch sits.

The unrestricted access ChatGPT Plus users recently enjoyed is now over. The strict five-hour usage limit for Codex and Work features is officially back as the company wants to stop users from draining resources.

Taiwanese furniture hardware maker King Slide Works has become an unlikely beneficiary of the AI boom, supplying specialized rails for data-center servers. Its shares have surged nearly 280% this year, helping its founder become Taiwan’s richest person.

AI video generators are significantly harder to use for free than typical LLMs like ChatGPT, with most providers being stingy with free credits. Getting them to create exactly what you want — while also maintaining a good, consistent look — is even harder. So I decided to test the alternative: hosting a local model such as MiniMax H3.

Microsoft is testing a new Windows 11 feature that lets users reserve portions of unified memory for graphics and AI workloads. It could be especially useful on systems such as AMD Strix Halo and Nvidia RTX Spark, where very large shared-memory pools can already be allocated to the integrated GPU.

Xiaomi has unveiled the AI Cube engineering prototype, a mini PC that combines its Xring O3, O100 and D100 chips for local AI model deployment. The system supports 120B and 3B models and can deliver up to 150 W performance.

Taiwanese prosecutors have indicted a senior Nvidia manager over his alleged role in a scheme to smuggle advanced AI servers to China. The case involves 74 Nvidia B300-equipped systems and could result in a five-year prison sentence for the alleged central figure.

You switched off "Improve the model for everyone" under Data Controls and assume that settles it. The toggle does exactly one thing, and only from the moment you flip it. Memories, connected apps, uploaded files and the safety checks carry on regardless. One gesture even sends a whole chat into training anyway.