Sep 26 – Oct 3, 2026

This week in AI infra

Every deprecation, API change, pricing shift, and model update that could affect builders shipping on AI APIs -- risk-classified and ready to share.

High-impact changes

11 changes across 5 providers.

0 critical11 high0 medium0 low
OpenAI (2)Microsoft (1)Apple Insider (1)Computerworldin (1)Tekedia (1)
highMicrosoftModel Release10/2/2026

Microsoft Digital Defense Report: How AI is reshaping the cybersecurity landscape

Microsoft is hardening model behavior and security boundaries, which matters for teams testing prompt injection resistance, browser safety, and agent reliability.

Action: Benchmark candidate model behavior before adopting in production.

highOpenAIModel Release10/2/2026

A model guide for the GPT-6 family

Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.

Action: Benchmark candidate model behavior before adopting in production.

highApple InsiderPricing Change10/2/2026

M6 Mac mini review: Excellent hardware badly hampered by too-small SSD

The M6 Mac mini has a faster chip and is solid hardware, but that $899 hurts and there is just not enough storage at that entry price for Apple Intelligence and all of your apps. M6 Mac mini review I have loved the Mac mini since day one, more than 20 years ago. There has been one in my house consistently since the first G4 model, generally attached to a television, or used as a server, and so forth. Oddly, the 2026 M6 Mac mini is not the cheapest Mac overall. The MacBook Neo has something to say about that, but that's just not as convenient for my own workflows. Continue Reading on AppleInsider | Discuss on our Forums

Action: Recompute cost guardrails and routing thresholds.

highComputerworldinPricing Change10/2/2026

Trump’s Super Intelligence edict supercharges .si domain registrations

US President Donald Trump’s rebranding of artificial intelligence (AI) as super intelligence (SI) has already caused a flurry of activity for one internet domain registry. SI is the international country code for Slovenia, and in the 24 hours following Trump’s executive order, Slovenia’s .si domain registry recorded 9,780 new .si registrations, compared to just 3,515 in the whole of the preceding month. Registrations were running at 2,726 per month and 110 per day around this time last year . Until now, registrars typically charged just $10 for .si domain registrations. The price is higher in the .ai domain belonging to the island of Anguilla: domain names there currently fetch $90. Conspiracy theorists are already wondering if there’s a link between First Lady Melania Trump’s Slovenian origins and the decision to rename AI as SI.

Action: Recompute cost guardrails and routing thresholds.

highTekediaModel Release10/2/2026

Huawei Launches Mate 90 Series, Putting Homegrown Chips and HarmonyOS at the Center of Its Smartphone Fight

Huawei Technologies is leaning harder on its homegrown chip architecture and operating system with the launch of its Mate 90 smartphone series, as the Chinese technology giant tries to sustain its premium handset comeback while U.S. restrictions continue to constrain access to advanced semiconductors. The new Mate 90 series, unveiled Thursday, represents another test of […] The post Huawei Launches Mate 90 Series, Putting Homegrown Chips and HarmonyOS at the Center of Its Smartphone Fight appeared first on Tekedia .

Action: Benchmark candidate model behavior before adopting in production.

highOpenAIModel Release10/2/2026

Has the time arrived for the retail chief AI officer?

Back in February 2018, Computer Weekly asked the question, “What are the retail jobs that don’t exist yet?”. In the midst of rapid technological change impacting the industry, it was a future-gazing piece speculating what roles retailers might soon be looking for considering the ongoing digital evolution, advancement in tech sophistication and changing customer demands. Then Missguided chief technology officer (CTO) John Allen was quoted predicting how automation and artificial intelligence (AI) best practice in retail will result in machines applying general processing rules, but with people on hand to intervene to make more specific commercial decisions when needed. That vision has become a reality retail is now embracing. But instead of the mooted AI intervention officer job role , we’re seeing fully fledged chief AI officers (CAIOs) enter the fray. Mohsen Ghasempour, an experienced technology leader with a PHD in computer science and CAIO at B&Q parent company Kingfisher Group, was announced as Morrisons’ first CAIO in September. Ghasempour remains very much in the minority as someone in retail with that job title – although US retail chain Target appointed Chandhu Nair as its first CAIO in August, and there is a groundswell of support and potential commercial necessity for others to join him. In his newly created role at Morrisons, Ghasempour is tasked with establishing a central data science and AI function for Morrisons that will take overall responsibility for data and AI strategy. In a varied job spec, the focus of the function at the grocer will be on customer growth, empowering colleagues, operational efficiency, and manufacturing and commercial excellence. He reports directly to the CEO, Rami Baitiéh, who is a vocal advocate of finding ways to use AI to drive business success. The line of reporting to the top shows how seriously the supermarket chain is taking the AI era. Baitiéh said the appointment reflects Morrisons’ commitment to “embracing the opportunities in data and AI at pace, with focus and in the right way”. Ghasempour himself spoke of helping the business “make effective and responsible use of AI in the years ahead” that, he hopes, will deliver real benefits for customers and colleagues alike. Responsible AI The use of the adjective “responsible” by Morrisons’ new recruit is pertinent against the current political and economic backdrop. US president Donald Trump met with tech company leaders at the end of September to agree voluntary AI safety standards. LinkedIn research released the same week shows the number of AI safety-related roles listed on its pages had grown by 91% over the previous year. Meanwhile, the Australian government set up a taskforce in September to review how it responds to AI-related cyber incidents , after it emerged an OpenAI agent gained unauthorised access to a Medicare website run by Services Australia. On 29 September, Computer Weekly caught up with a retailer with “responsible AI” in his job title. The meeting place was the Mach Alliance’s Mach X event in Amsterdam, which ran for two days and brought together retailers and other merchants with a raft of tech companies promising to bring composable, connected and open tech products to market. Just two years ago, few retailers would have had a senior executive with “responsible AI” in their job title. Today, as AI moves from experimentation into the heart of retail operations, roles such as that held by Pratyush Sinha are beginning to emerge. Sinha, who has just left his position as senior director and head of responsible AI at sports equipment retailer Decathlon, told this publication traditional technology leaders have historically been concerned with infrastructure, systems, deployment and cost. But autonomous AI systems are increasingly having a direct influence on organisational outcomes. “Agentic systems are in equal parts competing with infrastructure,” says Sinha. “The role of CIO was to get the most stable, confident and cost-efficient infrastructure in place that serves the company’s outcomes. Now agentic systems are doing much of this.” His conclusion is that the traditional technology hierarchy may itself need to change. “The CAIO role is more relevant in some organisations now than having a CIO or CTO or chief data officer,” says Sinha. Read more about retail technology Some 90% of retailers planning to boost spending on artificial intelligence to optimise e-commerce operations, as new research identifies different AI shopper personas and The Delivery Conference provides forum for debate. Ann Summers’ technology and supply chain director Jeannette Copeland talks through lessons learned during the retailer’s recent ESB overhaul . Sinha, who followed Barbara Martin Coppola from Ikea to Decathlon in 2022 when she moved from chief digital officer at the furniture retailer to take on the CEO role at the sports chain, said he has long considered “algorithmic accountability” in his work. The principle is relatively straightforward, he notes, saying if an algorithm is going to be deployed in a product, a retailer needs to understand how reliably it works and what happens when it encounters the messy complexity of the real world. “For me, what it means is that everything is observed, everything is re-constructible and everything is explainable,” he says of the key aspects of his role as head of responsible AI. With increasingly sophisticated AI models, particularly large language models, understanding exactly why a system produces a particular answer can be extremely difficult. For Sinha, the answer is not to treat responsible AI as a separate ethical exercise bolted onto technology projects – it needs to become part of how products are designed, trained and evaluated. One major area of focus for Sinha at Decathlon was to build an AI-powered e-commerce search function, and he notes this is where there is a huge responsibility to avoid bias in the search function’s results. One role of the head of responsible AI is to identify the difference between harmful bias and relevant context, he explains. “Let’s say a brown man, like me, [inserts into a search or chat function] that ‘I’m a brown man and I need shoes’ and the search finds shoes that are brown in colour – that will be harmful bias,” he notes. “But if a pregnant woman comes and says, ‘I’m pregnant and I need shoes’ and the model ignores the fact that there is pregnancy, that will be missing the context. So, knowing the difference between harmful bias and useful context is responsible AI.” Sinha says search systems need to be designed so they recognise these important factors in order not to upset customers or divert consumers to rival retailers. Changing responsibilities Sinha predicts that another emerging role in retail tech and wider business could be the “multi-agent architect”. The reason is because AI systems are now capable of doing more than simply responding to prompts or retrieving information. “Now AI can reason,” he says. “Now AI can exercise judgement. And now AI can look around the corner.” Agentic systems can, in his description, reason, make judgements and act – potentially coordinating several AI agents to achieve a particular business outcome. And architects will be – and already are, in his view – needed to orchestrate all of that in the form of a specific job role. Currently, he argues, the responsibilities for this new function sit somewhat “under the radar”, spread across existing technology jobs. As agentic systems become more deeply embedded in businesses, however, he expects the need for dedicated responsibility to become clearer. Indeed, roles and responsibilities of the tech, engineering and security teams was a hot topic throughout the Mach X event. Debating whose role is it to articulate any security consequences of AI deployment and what the rules are regarding AI usage, Simon Edwards, cyber security architect at shirtmaker and formalwear retailer Charles Tyrwhitt, called it a “shared responsibility”. “It’s never us as a single function,” he explained. “This is a new skill the business needs to adopt and a new responsibility it needs to adopt.” Neil Bennett, chief information security officer at the Post Office, who joined Edwards in a panel debate on the subject, added: “It’s a combination – it’s partly us, its partly legal, it’s partly DP [data protection], it’s a combination of different people who have a stake in this and a combination of people who have a detailed understanding.” Blending old and new Another common theme discussed in Amsterdam was the importance of finding the right blend between the new and the old guard. Fresh young talent leaving university, already tooled up with AI usage knowledge and bursting with ideas on how to implement it into the businesses they are joining, is now mixing with long-term IT practitioners who have grown up with the importance of risk management, guardrails and pragmatism when it comes to tech deployment. There’s room for both types in the industry, according to panellists debating the new agentic ecosystem. Travis Hess, CEO at e-commerce platform provider Commerce (previously known as BigCommerce), said: “Thinking of human capital and what’s the most important attribute in the world of agentic – it’s curiosity. “Those who are innately curious tend to do really well, but that also needs to be balanced with pragmatism.” Andrew Clarson, digital marketing and commerce practice leader at Deloitte Digital, who chaired the discussion, added: “The level of experience people have now has been normalised because everything is brand new. “The graduate coming out of university today is hungry and enthusiastic, and if they have the right personal attributes they can be just as – perhaps more – successful as someone with five years of actually increasingly irrelevant experience.” Pace of change The panel, which also featured Charles Tyrwhitt CTO Doug Wooten and MongoDB director of industry solutions Genevieve Broadhead, reflected on the risk to the retail industry of the pace of change in tech and the dangers of moving too fast with agentic commerce before relevant guardrails are implemented. Wooten said he was confident the curious nature of the graduates coming into the industry will dovetail neatly with the more pragmatic characteristics typically associated with those who have worked in IT for many years. Each generation helps the other, he noted. Broadhead suggested the new workforce generation coming through are not thinking about developing a skill and living with it for their whole career like their predecessors might have done. “They are thinking, ‘I’ll use this for six months then I’ll learn something new, something new, something new’,” she added, noting a person’s ability to adapt and to change – and to not be scared of learning new things – will be a vital skill in the agentic age.

Action: Benchmark candidate model behavior before adopting in production.

highStamford Latest NewsModel Release10/2/2026

Tokayev Urges AI-Driven Industrial Transformation, Proposes ‘Washington Model’ for Global AI Governance

The Astana Times provides news and information from Kazakhstan and around the world. ASTANA – President Kassym-Jomart Tokayev has called for a comprehensive AI-driven transformation of Kazakhstan’s industrial sector, urging the government and businesses to focus on measurable gains in productivity, technological modernization and high-value production. Chairing the third meeting of the Council for the Development of Artificial Intelligence on Oct. 2, Tokayev said AI was becoming a… The post Tokayev Urges AI-Driven Industrial Transformation, Proposes ‘Washington Model’ for Global AI Governance appeared first on The Astana Times .

Action: Benchmark candidate model behavior before adopting in production.

highStiri Pe SurseModel Release10/2/2026

BNR's Badea: Technology development and implementation of AI models no longer optional for emerging economies

The development of technology and implementation of artificial intelligence (AI) models are no longer optional for emerging economies, but fundamental elements of the new development model, as they need to avoid the middle-income trap, First Deputy G...

Action: Benchmark candidate model behavior before adopting in production.

highGoogleModel Release10/2/2026

Gemini 4 Argon Rolling Out Soon to Google AI Ultra Users

Google’s Gemini 4 Argon has emerged as a standout in artificial intelligence, combining advanced multimodal capabilities with robust reasoning performance. Currently in a limited release phase, the model is being tested by select cybersecurity partners and participants in Google DeepMind’s Fairwind program. One of its most notable features is its expanded token capacity, supporting up […] The post Gemini 4 Argon Rolling Out Soon to Google AI Ultra Users appeared first on Geeky Gadgets .

Action: Benchmark candidate model behavior before adopting in production.

highInternational Business TimesModel Release10/2/2026

Apple Reportedly Building AI Home Security Camera That Won't Record Video as Smart Home Hub Launch Nears

Apple is reportedly developing an AI-powered home security camera that would analyse activity and send text descriptions instead of recording video, according to Bloomberg's Mark Gurman.

Action: Benchmark candidate model behavior before adopting in production.

highBgrPricing Change10/2/2026

Paul Reubens Played A Sentient Spaceship In This '80s Sci-Fi Cult Classic Movie

You might recognize Pee-wee Herman's voice in this '80s sci-fi movie, but you won't find Paul Reubens' name in the credits. That's exactly how he wanted it.

Action: Recompute cost guardrails and routing thresholds.