top of page

AI Insights: Key Global Developments in September 2026

Staff Correspondent
12 minutes ago
7 min read

Welcome to the September 2026 ed. of our global AI update.


The past few weeks have brought some pretty big changes across the AI space, from OpenAI launching GPT-6 Astra to Apple finally giving Siri a much bigger AI upgrade. But beyond the big model launches, what really stood out was how AI is becoming part of more of the things we already use every day.We’re seeing more AI run directly on our devices, assistants getting better at handling actual tasks, and companies finding new ways to use AI outside the usual chatbots and coding tools.


Here are the key developments worth knowing.

OpenAI Launches GPT-6 “Astra” 


OpenAI has announced GPT-6 “Astra,” its most powerful AI model so far. 


OpenAI calls Astra the “world’s most intelligent and aligned model,” and says it has achieved top results in areas like professional work, coding, and reasoning. Astra is already rolling out to enterprise customers through ChatGPT Enterprise, Azure OpenAI, and AWS Bedrock, and will soon be available across all ChatGPT plans and through the API. One of the interesting things about Astra is that it can work with familiar software and web apps without needing a separate API for each one. 


Basically, businesses may not need a different AI tool for every task anymore.


For businesses, this could be a pretty big deal. Astra can help with data analysis, coding, research, scientific work, and other complex tasks, while also giving more accurate answers. Early testers say it makes fewer mistakes and handles things like maths and reasoning more reliably. 


OpenAI also says Astra can get work done using fewer tokens and fewer retries, which could help companies save on AI costs. This means businesses could use it for things like customer support, research, and software development, and automate tasks that usually take a lot of time. OpenAI has also focused on making Astra more reliable and reducing hallucinations, which is especially important when companies are trusting AI with real work. 


In simple terms, GPT-6 Astra is trying to make AI more powerful, more reliable, and much easier for businesses to actually use.

Source - OpenAI


Chinese startup Z.ai (formerly Zhipu AI) has released GLM-5.3-Flash, a 320-billion-parameter AI model that it says can compete with some of the top AI models from the West. 


The model was announced in late August, following the company’s earlier GLM-5.3 release. GLM-5.3-Flash can handle text, code, and conversations, along with multimodal tasks, and its weights will be made available for research. It is also designed to run efficiently on China’s home-grown “Shenwei” supercomputers.” This is another sign that China is working to build AI systems that don’t depend too heavily on foreign technology.


It adds another strong player to the global AI race. Chinese AI companies are showing that they can build powerful models outside the usual US-based AI ecosystem. GLM-5.3-Flash is built with China’s own infrastructure in mind and is planned to be open-source, giving local developers more freedom to use and improve it. Developers could use it for Chinese-language chatbots, business applications, or their own AI research without depending on foreign AI services. With trade and technology tensions still going on, having a strong home-grown AI model also gives China more control over its own AI ecosystem. And once the model is open, researchers around the world can also dig into it, test it, and see what it can really do.

Source - Z.AI


Stripe has announced that it is acquiring OpenRouter, an AI platform that lets developers switch between more than 400 AI models from 80+ providers.

The deal values OpenRouter at $7.5 billion, making it one of the biggest acquisitions in AI infrastructure so far.


So, why is Stripe interested in it? Well, companies don’t always want to use the same AI model for every job. One model might be faster, another might give better answers, and another might simply cost less. OpenRouter helps businesses automatically send each request to the model that makes the most sense. By bringing this together with Stripe’s payment system, companies could also have an easier way to track their AI usage, manage costs, and handle billing. 


Basically, instead of picking one AI model and sticking with it, businesses can let the system pick the right one for each task, while Stripe takes care of the money side.

Source - Stripe


At IFA 2026 in Berlin, NVIDIA showed off a bunch of new “local AI” tools that could bring powerful AI models straight to your PC. 


One of the biggest announcements was NVIDIA PAIR, or “Personal AI Router,” which can split AI tasks across multiple GPUs at home to make them run faster. NVIDIA also introduced the upcoming RTX Spark Windows PCs from Lenovo and Acer, built specifically for AI-heavy tasks. 


The company also announced updates through llama.cpp and vLLM that could make local AI models run up to 1.9x faster. It is even working with game studios to bring games to RTX Spark PCs.


AI may not always need the cloud anymore. More powerful PCs could soon run advanced AI models right on the device, whether it’s for voice assistants, gaming, or editing videos. That also means sensitive information can stay on your computer instead of being sent to a remote server. NVIDIA says models such as China’s Nemotron-3.5 Lightning and Z.ai’s GLM-5.3-Flash are also being tuned to work with its local hardware. For users, this could mean faster responses and more privacy. For developers, it could mean lower cloud bills and less waiting around for servers. Basically, AI is slowly moving from giant data centers to the device sitting right in front of you.

Source - NVIDIA


On September 14, Apple launched “Siri AI” with iOS 27, giving its voice assistant a major upgrade. 


Powered by Apple Intelligence, the new Siri can use information from your messages, emails, and even what’s on your screen to give more useful answers. Apple calls it an “entirely new version of Siri” that is more personal, capable, and conversational. For example, Siri can now schedule an event based on an email, or look through the web and your apps at the same time to find an answer. And yes, Apple is still putting privacy first, with much of the processing happening on the device. The rollout starts in English, with more languages coming later.


This is a big step for Apple’s AI plans. 


The old Siri was great at things like setting alarms or calling someone, but the new version is meant to handle much more complicated tasks. It can help with scheduling, summarising content, editing, writing, and even keep track of what you were talking about during a conversation. Apple is using a mix of on-device and cloud processing to bring these AI features without giving up its focus on privacy. The launch also puts Siri in the same race as AI-powered assistants from Google and Amazon. Apple is taking a careful, phased approach with the rollout. Siri is no longer just about “Hey Siri, set an alarm.” It wants to become an actual AI assistant.

Source - Apple


Google has quietly rolled out “Gemini 3.5 Transcribe,” a new speech-to-text model built to make transcription more accurate and better at understanding context. 


It can handle things that usually mess up transcripts, like background noise, industry jargon, and all those little “umms” and “ahhs” we make while talking.


Google says the model is already being used for new voice features, including “Rambler” on Gboard, and developers can also access it through Google AI APIs.


The update may sound small, but it can be pretty useful. Businesses and developers can use better speech recognition to build voice assistants, captions, meeting tools, and analytics apps that actually understand how people speak. Gemini 3.5 Transcribe supports more than 85 languages and can automatically add punctuation and capital letters, so the final text doesn’t look like one giant sentence. It can even turn spoken commands into actions, such as generating an image or pulling up information. Basically, Google is trying to make talking to AI feel less like giving a robot commands and more like having a normal conversation.

Source - Google


Google has teamed up with the UK government on “Operation Blue Skies,” a project that uses AI to help make air travel a little better for the planet. 


The idea is pretty interesting: AI can help airlines choose flight routes that avoid creating contrails, those white lines in the sky made of ice clouds that can trap heat. By looking at weather and flight data, the system can suggest small changes to a plane’s route or altitude. The UK government says the project could help airlines reduce contrails by up to 50%.


Why does this matter? Aviation creates a lot of emissions, and this project looks at a way to reduce some of that impact without completely changing how planes fly. Google’s AI models can work with the UK’s air traffic and weather data to find routes that are better for the environment while still getting passengers where they need to go.


The first trials are focused on North Atlantic routes, but if the results are good, the idea could be used in other parts of the world too. It’s a pretty cool example of AI solving a problem that has nothing to do with chatbots or coding, but with something as big as the climate.

Source - Google

It feels like we’re getting closer to a point where these tools are simply part of the technology we use every day. 


Assistants are doing more than just answering questions, models are handling harder tasks, and we’re seeing new uses in places like phones, laptops, aviation, and scientific research. At the same time, companies are working on making these systems faster, safer, and more private. There’s still a lot to figure out, but it’s clear that the way we use technology is changing pretty quickly.


Stay tuned for our next issue, where we’ll cover developments in model deployment, risk management, and global policy. As always, we welcome your feedback or tips on stories to include. Feel free to reach us at info@riskinfo.ai.


Comments

bottom of page