Smartwatch AI spots signs of high blood pressure without an arm cuff

✕Sun 8 Nov ‹›

Health research · Sun 8 Nov

Smartwatch AI spots signs of high blood pressure without an arm cuff

65.8%of high blood pressure cases the watch caught (first clinic check: 55.3%)

Google Research says an AI system can spot signs of high blood pressure from an ordinary smartwatch's pulse and motion sensors, with no arm cuff needed to set it up. The peer-reviewed study was published in Nature on 5 Nov 2026. It flags who should get a proper check; it does not give you a blood pressure number.

Key points

How it works: the watch's light sensor tracks blood flow at the wrist, and its motion sensor picks up tiny jolts from each heartbeat. The AI averages these over several days and adds your height to estimate the chance you have high blood pressure.
The test: 196 adults with no known heart disease and no blood pressure medicine wore a Google smartwatch for seven days. Results were checked against 24-hour blood pressure monitoring, the method doctors treat as the gold standard.
Results (from the Nature paper): the watch caught 65.8% of people who had high blood pressure, versus 55.3% for a first clinic reading. Both correctly cleared 90% of people who did not. Nature's press release says the gap was , so read it as 'about as good', not 'better'.
Limits: the watch still missed about a third of cases. Google calls this early research, not a current or planned product feature, though it says the work helped inform blood pressure trends in the Google Health app's Health Guardian feature.

Why it matters: More than half of adults with high blood pressure do not know they have it. If a watch ever nudges you, confirm with a proper cuff reading and a doctor, and remember a quiet watch is not an all-clear.

Sources (3)

Detecting signs of hypertension using wearable sensors · Google Research
Opportunistically Detecting Signs of Hypertension on a Consumer Smartwatch · Nature
Technology: Smartwatches that can clock high blood pressure · Nature Portfolio press release

Next · How companies use AIJapan's freee says its own small AI model cut costs to 1/50 →

✕Sun 8 Nov ‹›

How companies use AI · Sun 8 Nov

Japan's freee says its own small AI model cut costs to 1/50

1/50of the cost freee reports versus a big cloud AI model

Japanese accounting software company freee reads about one million receipts, invoices and other business documents a month with AI. Instead of sending them all to a big general AI model, it built its own for one job: picking out details like the date, amount and supplier. The figures are freee's own, from a talk at AWS Summit Japan 2026 that ITmedia reported on 7 Nov 2026.

Key points

Company: freee, Japan, accounting and back-office software. Problem: big AI models read documents well, but at a million documents a month, paying per request cost too much.
What they built: an 8-billion- given extra training on one task () with about 10,000 checked examples, plus tens of thousands more for testing. It runs on freee's own servers in AWS. A separate step still turns each image into text first.
Measured results (company's own figures): up to 10 points more accurate on text details such as description and supplier, 5 times faster, and 1/50 of the cost, compared with a Claude Sonnet 4.6-class model used through a cloud service. freee says the saving comes mainly from paying for server time instead of per request.
Lesson for smaller teams: this pays off only with steady, high volume and one narrow task. freee's rule of thumb is that if a small model like Claude Haiku 4.5 already does the job reliably, a self-run small model may compete. Measure real speed early, and keep testing after launch.

Why it matters: AI bills grow with every request. For one repetitive job at big volume, a small model you run yourself can be far cheaper, but someone has to test and maintain it.

Sources (1)

freeeは「AIコスト問題」をどう乗り越えた？月100万件規模の処理で、汎用LLM比50分の1に (How freee beat its AI cost problem; translated) · ITmedia Enterprise

Next · Your AI subscriptionJudge says Claude Max's '20x more usage' ads may mislead buyers →

✕Sun 8 Nov ‹›

Your AI subscription · Sun 8 Nov

Judge says Claude Max's '20x more usage' ads may mislead buyers

6xMax 20x weekly use vs Pro, by the subscribers' maths for one model

A US federal judge said on 6 Nov 2026 that Anthropic's promise of "5x or 20x more usage than Pro" for its Claude Max plans is "plausibly misleading". Key parts of a by two subscribers can go ahead. Nothing has been proven yet: at this early stage, the court treats the complaint's claims as true.

Key points

The plans at the time: Pro cost about RM69 to RM81 (US$17 to US$20) a month, Max 5x about RM406 (US$100) and Max 20x about RM811 (US$200).
The dispute: Max gives 5 or 20 times Pro's use inside each five-hour session, but the is not 5 or 20 times higher. The subscribers calculate that for one Claude model, Max 5x gets about 3.5 times Pro's weekly use and Max 20x about 6 times.
The ruling: Magistrate Judge Alex Tse kept one subscriber's consumer-protection claim, plus claims of negligent misrepresentation and breach of contract. He dismissed the others but let the subscribers refile them. No class has been approved and there is no settlement.
Anthropic had asked for the whole case to be thrown out, arguing it promised more use per session and said weekly limits apply. An amended complaint is due by 2 Dec 2026, according to the case tracker LimitHit.

Why it matters: "More usage" can mean per session, per week or per month. Before you upgrade any AI plan, check the weekly limit, not just the headline multiplier.

Sources (3)

Anthropic must face claims its Claude Max pitch misled users · Courthouse News Service
Order on motion to dismiss, Case 3:26-cv-05763-AGT · US District Court, N.D. California
Claude usage limit lawsuits (2026): what's alleged and where they stand · LimitHit

Next · AI at workGitHub rebuilt its storage so 5,000 AI agents can share one project →

✕Sun 8 Nov ‹›

AI at work · Sun 8 Nov

GitHub rebuilt its storage so 5,000 AI agents can share one project

7,200updates a second to one project in GitHub's own test

GitHub says it rebuilt how it stores code so thousands of AI coding can save work to the same project at once. In its own tests, 5,000 agents sent 7,200 updates a second to one project while reading stayed fast. These are GitHub's internal test figures, from its engineering blog on 5 Nov 2026.

Key points

Why: AI agents save changes far more often than people do. GitHub says a busy project used to mean dozens of updates a second, so it aimed for at least 100 times more.
The idea: most of a code update can be stored without waiting for anything else. Only the tiny step that records which version is now the latest needs everyone to agree, so GitHub gave that step its own fast service.
For people using GitHub, nothing changes: you save and fetch code the same way, with the same protections. Behind the scenes, nearly all projects now sit in Microsoft's Azure storage.
Limits: the 5,000-agent figure comes from GitHub's own testing, not an independent check, and the post does not give a date for when every project moves to the new system.

Why it matters: Big tools are being rebuilt for many AI agents working on the same files. If your team tries coding agents, plan time for people to check what the agents changed.

Sources (1)

Inside GitHub's new Git storage architecture · The GitHub Blog

Next · Southeast AsiaU Mobile opens its 5G-A sandbox to Cradle startups for AI tests →

✕Sat 7 Nov ‹›

Southeast Asia · Sat 7 Nov

U Mobile opens its 5G-A sandbox to Cradle startups for AI tests

5G-Asandbox U Mobile is opening to Cradle-linked startups

Malaysia's U Mobile is deepening work with so local startups can try and AI ideas on a real . The pair held demo days and roadshows, including one at U Mobile's Hub on 28 October. Funding and commercial deals are not part of the offer.

Key points

What is on offer: U Mobile's Enterprise Innovation Platform gives eligible startups a place to test and show 5G-A and AI tools to businesses. Cradle brings programs, ecosystem links and possible later funding paths. Access still depends on technical fit.
Recent events: Since September 2026 the partners ran a Cradle roadshow in Sarawak and the MYStartup Pre-Accelerator Demo Day at the EIP Hub. U Mobile's network ratings (Ookla, Opensignal) are company-cited speed and reliability claims.
What is not promised: TechNode Global and e27 both stress that cheques and paid contracts are not automatic. Startups get time and industry meetings first.
Why now: U Mobile's CTO Woon Ooi Yuen said startups need a place to test against real business needs. Cradle's CEO Norman Vanhaecke said the goal is helping Malaysian firms grow beyond the home market.

Why it matters: Local builders get a rare chance to try AI on a live 5G-A network without buying one. Treat it as a test bed, not free funding.

Sources (2)

Next · How companies use AIIndia's cult.fit says Codex cut support costs by more than half →

✕Sat 7 Nov ‹›

How companies use AI · Sat 7 Nov

India's cult.fit says Codex cut support costs by more than half

50%+drop in support costs cult.fit reports after AI agents

Fitness chain cult.fit standardised on OpenAI's coding , then let non-engineers build tools with it. The company's own figures in an OpenAI customer story claim big savings and many new internal apps. Numbers below are vendor case study claims.

Key points

Company: cult.fit, India, about 708 gyms in 77 cities and nearly one million paid members. Industry: fitness and consumer products.
What they built: one shared Codex setup for engineering (including SRE and security agents), then office hours so HR and other teams could ship apps. One engineer rebuilt an internal dashboard over a weekend after a software price jump.
Measured results (company's own figures via OpenAI): more than 20 internal apps from non-engineers; over about RM405,000 (US$100,000) a year saved on that dashboard; more than 50% lower customer support costs; a social feed shipped in two months versus an earlier estimate of more than a year.
Lesson for smaller teams: pick one coding agent and train people around it, then measure support and software bills before and after. Treat the savings as company claims until you see your own numbers.

Why it matters: A large Asia fitness firm is using the same coding agent for engineers and office staff. The savings are company claims, but the pattern of standardising on one tool is easy to copy.

Sources (1)

Cult.fit builds operational muscle with Codex · OpenAI

Next · How companies use AIGreenlight's Luna clears over 90% of security cases on its own →

✕Sat 7 Nov ‹›

How companies use AI · Sat 7 Nov

Greenlight's Luna clears over 90% of security cases on its own

90%+of monthly security cases Greenlight says Luna closes alone

US family-finance firm Greenlight built Luna, a security helper on . AWS says Luna resolves more than 90% of up to 4,000 security and risk cases a month without a person. Figures are from an AWS customer post.

Key points

Problem: 3,000 to 4,000 alerts a month from dark-web card leaks, odd logins, infrastructure tools and fraud signals. A human case could take 15 to 30 minutes.
What they built: a Tier-1 supervisor that calls more than 30 specialist , plus a Tier-2 deep dive. It runs on Bedrock AgentCore runtime and memory inside a private network.
Result (AWS / company claims): more than 90% of cases closed without a human; routine in single-digit seconds. Complex cases still go to analysts with full .
Lesson: agent stacks help when the work is high volume and rule-heavy. Keep a human path for the hard cases, especially in finance.

Why it matters: Banks and fintechs face the same alert flood. An autonomous first line only works if someone still owns the tricky calls.

Sources (1)

How Greenlight Financial built an autonomous SOC solution on Amazon Bedrock AgentCore · AWS Machine Learning Blog

Next · China AIKimi leads 89% of open-source AI spend by companies, study says →

✕Sat 7 Nov ‹›

China AI · Sat 7 Nov

Kimi leads 89% of open-source AI spend by companies, study says

89%of open-source AI company spend Menlo says went to Chinese models

A study, reported by SCMP, says Chinese open models took about 89% of global enterprise spending on AI. Moonshot's Kimi led with 61%, ahead of DeepSeek and Z.ai. Proprietary US models still take most overall AI revenue.

Key points

Shares (Menlo Ventures via SCMP): Kimi 61%, DeepSeek 15%, Z.ai GLM 11%, Meta's Llama about 7% of open-source enterprise spend.
Caveat: open-source is still a small slice next to paid closed models. The ranking is about that niche, not the whole AI market.
Related money: Beijing video startup Sand.ai closed about RM809 million (US$200 million) led by CoStone and Wangsu, taking 10-month fundraising to about RM1.21 billion (US$300 million) and a company-claimed above about RM4.05 billion (US$1 billion). Its MAGI-2 Preview is a 114-billion- video model, the company says.
Also at Wuzhen: China showed an still in pilot, and about 20 firms including Huawei, Alibaba and Tencent signed an agent .

Why it matters: Chinese open models are cheap for companies to try. Closed US tools still take most of the money, so both tracks matter.

Sources (3)

China's AI models, led by Kimi, dominate open-source enterprise spending · SCMP Tech
Sand.ai closes $200 million Series A+ for video AI · TechNode Global
China Unveils an AI Agent Identity Registry at Wuzhen as 20 Organizations Sign Self-Discipline Pact · Pandaily

Next · AI video toolsGoogle's Gemini Omni 1.2 Flash makes 15-second clips cheaper →

✕Sat 7 Nov ‹›

AI video tools · Sat 7 Nov

Google's Gemini Omni 1.2 Flash makes 15-second clips cheaper

15smaximum native clip length in Gemini Omni 1.2 Flash

Google launched 1.2 Flash for the Gemini and business plans. Clips can run up to 15 seconds, up from 10, and the price falls about 30% to about RM0.28 (US$0.07) per second at 720p. You can draft cheap, then upscale the keepers.

Key points

New limits: native video up to 15 seconds, with length set in one-second steps from 3 to 15. Multi-turn edits let you fix one detail without remaking the whole clip.
Price: about RM0.28 (US$0.07) per second at 720p, Google says 30% less than Omni 1.1. Draft at 360p or 720p, then upscale to 1080p or 4K.
Consistency: you can mix character, product or style stills, short motion clips and audio so faces, packaging and brand colours stay steadier across angles.
Related: Anthropic started a US beta of Claude Personal Finance, which links bank accounts through in read-only mode. It answers money questions but cannot move funds.

Why it matters: Shorter, cheaper video tools help small shops make product clips. Check the final cut yourself before you pay to run ads.

Sources (2)

Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips · Google AI Blog
Anthropic adds Claude Personal Finance bank linking in the US · ITmedia AI+

Next · Southeast AsiaSingapore leads Commonwealth youth AI readiness at 77.6 →

✕Sat 7 Nov ‹›

Southeast Asia · Sat 7 Nov

Singapore leads Commonwealth youth AI readiness at 77.6

77.6Singapore's Ayre Index score out of 100 for youth AI readiness

The new ranks Singapore first among 56 Commonwealth countries for how ready young people aged 15 to 29 are for AI, with a score of 77.6 out of 100. Australia is third. e27 says the harder gap is jobs and rights, not gadgets.

Key points

What it measures: technology access, digital skills, education, employment and safeguards for youth across the Commonwealth.
Singapore's mix: strong digital skill and confidence scores; weaker on employment/enterprise and rights, safety and participation, according to secondary coverage of the index.
Regional read: e27 argues Australia's third place shows that owning the tech is easier than building fair work and safety rules around it.
Also in the region: AMD, FuriosaAI and MangoBoost said they plan a Seoul AI cluster, a nearby Northeast Asia build-out that SEA buyers may watch for hardware supply.

Why it matters: Scores help governments compare homework. Young workers still need jobs and clear rules after the ranking photo is taken.

Sources (2)

Singapore tops a new Commonwealth AI readiness index, but Australia's results show where the real gap lies · e27
AMD, FuriosaAI and MangoBoost plan Seoul AI cluster · TechNode Global

Next · AI securityOpenAI's GPT-6 Cyber finds 81 Firefox bugs and 948 Linux exploits →

✕Fri 6 Nov ‹›

AI security · Fri 6 Nov

OpenAI's GPT-6 Cyber finds 81 Firefox bugs and 948 Linux exploits

81new high-severity Firefox bugs OpenAI says GPT-6 Cyber found

OpenAI launched GPT-6 Cyber, a model built to find and test software holes for security teams that are allowed to use it. OpenAI says it scored 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber, a 51% relative gain. Access starts with approved Red customers only.

Key points

Real systems: OpenAI researchers used it to develop working for 948 Linux bugs and to find 81 previously unknown high-severity Firefox bugs. Mozilla has begun shipping fixes. Cisco, Calif, CodeRabbit and others say they are using it for research and .
Staying in bounds: OpenAI trained the model to stay inside the task a researcher sets. In one test, it avoided attacking out-of- services in 99.93% of runs without a safety prompt, and 100% with the prompt.
Risk level: OpenAI rates the model Critical for cybersecurity under its , up from High for the previous Cyber model. It pairs the release with stronger misuse checks and a path for critical-infrastructure defenders.
Who can use it: GPT-6 Cyber starts on the and for a select Daybreak Red group, then expands. Daybreak Blue access for related defender tools is also widening.

Why it matters: Security teams may find bugs faster, but the same skills can be misused. Only approved users get the strongest tools, and ordinary readers should treat viral 'leak' claims with care.

Sources (3)

GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents · OpenAI
OpenAI announces cybersecurity model GPT-6 Cyber; ExploitGym score up 51% · ITmedia AI+
OpenAI's strongest cyber AI model GPT-6 Cyber; 81 new Firefox bugs · IThome

Next · Southeast AsiaAWS aims Amazon Quick at 445,000 officers; Microsoft picks Johor →

✕Fri 6 Nov ‹›

Southeast Asia · Fri 6 Nov

AWS aims Amazon Quick at 445,000 officers; Microsoft picks Johor

445,000government officers AWS aims to reach with Amazon Quick

Two big firms moved on Malaysia's AI plans on 5 November. AWS launched Semarak AI Malaysia to help the public sector, businesses and communities build with AI. Microsoft said its next Malaysia , Southeast Asia 3, will be in Johor Bahru.

Key points

AWS Semarak: three pillars cover government (AI Negara), business (AI Ekonomi) and communities (AI Rakyat). It builds on AWS's about RM29.2 billion (US$7.21 billion) Malaysia commitment.
Government tools: Malaysia is launching GovHub with AWS's coding tool. A pilot starts with 500 developers in 2026, aiming for 4,000 by 2028. AWS is also investing to put , its work assistant, in front of up to 445,000 government officers for tasks such as reading procurement data and policy research.
Small firms and schools: with GDEX's GD XCHANGE and VSTECS, eligible small businesses can get six months of free Quick and Kiro after an AI readiness check, aiming for 10,000 firms by 2028. More than 3,000 students and lecturers joined a polytechnic AI contest, AWS says.
Microsoft Johor: Southeast Asia 3 is part of Microsoft's about RM8.9 billion (US$2.2 billion) four-year Malaysia plan. Backup generators will use renewable . Microsoft says more than 734,000 people were trained under AIForMYFuture by October 2025, toward an 800,000 target by 2027. Digital Minister Gobind Singh Deo also urged that Google's RM61 billion plan create Malaysian tech, not only more computing.

Why it matters: Malaysia is stacking , cloud regions and AI tools for civil servants and small firms at the same time. The officer and training numbers are company or programme targets, so delivery still needs watching.

Sources (3)

AWS launches Semarak AI Malaysia, putting Amazon Quick tool to 445,000 government officers · TechNode Global
Microsoft to open Southeast Asia 3 cloud region in Johor Bahru, per $2.2B Malaysia plan · TechNode Global
Gobind: Malaysia must turn Google's US$15bil investment into homegrown AI innovation · Digital News Asia

Next · How companies use AIJapan's Nitori found AI analysis fails until people name their data →

✕Fri 6 Nov ‹›

How companies use AI · Fri 6 Nov

Japan's Nitori found AI analysis fails until people name their data

2,600main tables Nitori had to make AI-readable

Furniture retailer Nitori moved sales data into Google and ended up with about 2,600 main tables. Staff who already knew the data thrived; everyone else got lost. The fix was not a smarter chatbot. It was writing plain labels for what each column means.

Key points

The problem: with so many tables, only a few experts could still see the whole picture. Off-the-shelf chat tools failed because column names followed Nitori's own codes, such as DELVTO_CRP_KBN for 'delivery destination company type'.
The work: the IT team added meaning and into table and column properties, linked them to an internal glossary, and used small for specific jobs instead of one giant agent.
What changed: people without SQL skills can ask for figures in everyday language. New managers can learn the data layout faster. Gemini can also help rewrite old data-cleaning jobs once the labels exist.
Lesson from Nitori's Kei Kobayashi: you do not need every table perfect. Labelling the top few most-used ones is enough to unlock most of the gain.

Why it matters: Buying an AI chat box does little if your spreadsheet names only make sense to three people. Clear labels are the boring step that makes AI useful for the rest of the team.

Sources (1)

Nitori: how naming rules blocked conversational BigQuery analysis, and how labels fixed it · ITmedia Enterprise

Next · AI securityA fake AGENTS.md in Downloads can hijack coding agents with no click →

✕Fri 6 Nov ‹›

AI security · Fri 6 Nov

A fake AGENTS.md in Downloads can hijack coding agents with no click

0 clicksneeded for a website to plant the fake instruction file

Security firm Forever Security says it caught a new attack on real company computers. A website quietly drops a fake into the Downloads folder. Coding that trust that folder then rewrite their own settings and open the door to .

Key points

How it works: tools such as and Anthropic Claude Code read AGENTS.md or CLAUDE.md for instructions. Browsers can also save files to Downloads without a click. Together, that means an agent can treat a planted Markdown file like a program.
What the fake file did: one real sample told Codex to edit ~/.codex/config.toml and install a bad called function-color. A CLAUDE.md test rewrote .claude/settings.json so traffic went to an attacker's server about half the time.
How common: Forever Security says about half of its customer firms ran coding agents from Downloads, and about a quarter of agent sessions opened a browser.
Five free checks: do not run agents from Downloads; read unknown AGENTS.md or CLAUDE.md first; inspect config.toml and settings.json for strange servers; keep approval gates on; if you find a bad file, treat as leaked and reset them. Finding function-color installed is a red flag.

Why it matters: This is not a bug the vendors can simply . Keep agent work folders away from places the web can write to, and skim instruction files before you let an agent follow them.

Sources (1)

Fake AGENTS.md in the Downloads folder can hijack coding agents without a click · ITmedia Enterprise

Next · AI productsChatGPT Images 2.6 is 50% faster and can make GIFs →

✕Fri 6 Nov ‹›

AI products · Fri 6 Nov

ChatGPT Images 2.6 is 50% faster and can make GIFs

50%faster image generation vs Images 2.5, OpenAI says

OpenAI released ChatGPT Images 2.6 on 4 November for every ChatGPT and plan, including free. It says generation is about 50% faster than Images 2.5, edits keep character details better, and GIF animation is rolling out.

Key points

New tricks: users can ask for looping GIFs or turn a sketch into a waving character. Saved reference photos can be named once, then reused in later chats with an @ mention, so you do not re-upload the same dog or product shot.
For developers: the offers a faster Flare option and a higher-quality Sunburst option, plus SVG export and a complexity setting that controls how much the model plans before drawing.
Coverage: ITmedia and QbitAI both report the rollout. GIF support is still arriving in stages.

Why it matters: Faster edits and reusable references matter for small teams that make social posts and product shots. Treat GIFs as a convenience, not a reason to skip a human check before publishing.

Sources (2)

OpenAI announces Images 2.6: 50% faster, GIF animation support · ITmedia AI+
GPT-Image-2.6 launches fully; direct GIF and SVG output · QbitAI

Next · Malaysia and AIGoogle will invest about RM61 billion in Malaysia's AI data centres →

✕Thu 5 Nov ‹›

Malaysia and AI · Thu 5 Nov

Google will invest about RM61 billion in Malaysia's AI data centres

RM61 billionGoogle's planned Malaysia investment in 2027-2028 (about US$15 billion)

Google plans to spend about RM61 billion (US$15 billion) in Malaysia in 2027 and 2028, mostly on in Selangor and Negeri Sembilan. It announced the plan on 4 November with Prime Minister Anwar Ibrahim. It also promises training for 800,000 Malaysians and more clean power for the national grid.

Key points

What is being built: Google's first campus at Elmina Business Park in Selangor is already running. New campuses are planned in Selangor and in Port Dickson, Negeri Sembilan. They will also support Google's coming in Malaysia.
Jobs: Google expects the building work to support more than 37,000 jobs a year and add about RM4.5 billion (US$1.1 billion) a year to the economy. Once running, it says the sites will keep over 9,000 jobs that pay about 45% above the national median, with more than 80 Malaysian suppliers.
Skills: four programmes aim to train 800,000 Malaysians by 2031. They cover teachers and young people, small business owners, civil servants and people looking for work, starting in the host states.
Power and water: Google says it will add more than 1 of clean energy to the national grid with Tenaga Nasional, and pay for all the power its sites use. It is also funding river work in the Sungai Selangor basin and water tanks at four Selangor schools that can give 2,000 people water for up to two weeks.

Why it matters: This is one of the biggest tech investments in Malaysia this year. The jobs and training numbers are Google's own estimates, so it is worth watching whether the training reaches people beyond the host states.

Sources (2)

Next · How companies use AIMalaysia's U Mobile cut some support tickets by 84% with AI helpers →

✕Thu 5 Nov ‹›

How companies use AI · Thu 5 Nov

Malaysia's U Mobile cut some support tickets by 84% with AI helpers

84.5%fewer network and setup tickets at U Mobile, Jan to Sep 2026

Malaysian telco U Mobile built an AI that checks its systems and suggests a fix before staff act. A drug-industry data firm cut a 30-hour setup job to about 4 hours. But a large study warns that AI-written code adds checking work for people.

Key points

U Mobile: its AI helper, AiDaH, looks up the account, finds the cause and proposes a fix, and a person confirms before anything changes. From January to September 2026, monthly fell 84.5% for network and setup problems, 81.6% for general questions and 48.7% for billing, U Mobile says.
In one case it found a customer had tried the same purchase more than 30 times over three days, all blocked by one stuck order. Staff use it inside Microsoft Teams, so there was no new app to learn.
Indegene, which serves drug companies, uses an AI helper to write the checks that catch wrong or missing data, called . Setting these up for one dataset fell from about 30 hours to about 4 hours, and people every suggestion.
A Harvard study of more than 700 companies found AI coding raised the amount of code by 30%, but finished work by only 12%. Code review took 49% longer, because people had more AI work to check.

Why it matters: The clearest wins come when AI does the digging and a person makes the final call. Plan time for that checking, or the savings shrink.

Sources (3)

How U Mobile built agentic telco operations with Amazon Bedrock AgentCore · AWS Machine Learning Blog
How Indegene is modernizing and scaling enterprise data quality in life sciences · AWS Machine Learning Blog
Harvard study of 718 companies: AI coding agents and the code review bottleneck · InfoQ China

Next · Robots at workSix robots ran a Shanghai shop for 24 hours, doing 1,730 tasks →

✕Thu 5 Nov ‹›

Robots at work · Thu 5 Nov

Six robots ran a Shanghai shop for 24 hours, doing 1,730 tasks

1,730tasks done by six robots in one 24-hour shop shift

In China, are moving from stage demos into real shops, pharmacies and factories. The numbers come from the companies themselves, and some key results, like how often robots failed, are still missing.

Key points

AgiBot: six G2 humanoid robots ran a convenience store in Xuhui, Shanghai, from 10am on 2 November to 10am on 3 November. They greeted 683 customers, made 147 coffees, restocked shelves 277 times and did 1,730 tasks in all. Four of the six passed the company's .
AgiBot has not shared how often the robots failed or needed human help, and fine tasks with soft or small items were still hard.
Ant Group: its robots already pick medicines in several Guoda Pharmacy stores in Shanghai, moving through aisles just 80 cm wide without any change to the shelves. Its open robot models have been downloaded about 2.2 million times, the company says.
Xiaomi: a humanoid robot worked its first full 8-hour shift at a car factory sorting station. Xiaomi says success rates across three stations were 94% to 98%.

Why it matters: Robots that fit into existing shops and pharmacies, without rebuilding them, are the kind that could one day help in Malaysian clinics and stores. Until companies share failure rates, treat these numbers as demos, not proof.

Sources (3)

Next · AI and healthA son uses AI to help his mum through cancer →

✕Thu 5 Nov ‹›

AI and health · Thu 5 Nov

A son uses AI to help his mum through cancer

1 pagebrief that keeps the whole care team on the same facts

Creator Peter Yang's mother has cancer again. He shared the steps he follows with AI, mostly for paperwork and preparing questions, not for medical answers. He says online claims that AI has cured cancer are false.

Key points

Keep a one-page brief: patient details, the care team, next steps, key facts, simple explanations of medical words and a dated care log. He keeps it in a shared Google Doc and asks AI to update it.
Collect medical records early: written reports, scan files and tissue samples can take weeks to arrive. With all of them in one place, the AI answers from her real history, not general advice.
Research with care: he uses AI to learn the standard treatment, prepare for a and look through , but warns that some early trials can make things worse. for her will cost about RM16,300 (US$4,000).
The real work is still faxes and CDs of scans, he says. His last step is to protect time with family. He shares the method free as a '' that chatbots can follow.

Why it matters: For caregivers, AI's most useful job may be keeping track of paperwork and preparing good questions for short appointments. Treatment decisions still belong with doctors.

Sources (1)

My Mom Has Cancer Again. Here's How I'm Using AI to Fight It · Peter Yang

Next · Check before you shareA Claude limits rumour was false, and leaked 'model files' look fake →

✕Thu 5 Nov ‹›

Check before you share · Thu 5 Nov

A Claude limits rumour was false, and leaked 'model files' look fake

400+fake-IT attacks in Microsoft Teams counted by Arctic Wolf

Several AI stories spread online this week without proof. Anthropic denied one, and a fake download is a safety risk. Separately, hackers are posing as IT staff in Microsoft Teams.

Key points

On 30 October Anthropic renamed its top Claude plan Ultra, at the same price of about RM900 (US$220) a month. Users on X said usage was cut in half. Anthropic's product lead Mike Overell said no: the was always 10 times the Pro plan, and he apologised for the confusion.
A Hugging Face page claimed to share the of Anthropic's unreleased Mythos model. Researchers found signs the files were made up, and there is no reliable proof of a leak. Do not download files like these.
Posts also show test results said to come from an unreleased 'Claude Fable 5.5'. Anthropic has not confirmed it exists.
Security firm Arctic Wolf counted more than 400 attacks where hackers flood an inbox with spam, called , then message the person in Microsoft Teams as 'IT' and ask for . IBM's security experts say real IT staff will not ask for your password.

Why it matters: Before sharing an AI rumour, look for the company's own statement. If 'IT' contacts you out of the blue after a flood of emails, call your IT team on a number you already know.

Sources (3)

Claude plan rename halves usage? Anthropic denies: Max 20x weekly limit was always 10x Pro · ITmedia AI+
Claims of leaked Mythos weights and Claude Fable 5.5 tests lack proof · IThome
Did the Knicks' phishing test go too far? Plus, hackers pose as IT in Teams · IBM Technology

Next · Southeast AsiaMalaysia opens an AI-powered online centre for ASEAN startups →

✕Thu 5 Nov ‹›

Southeast Asia · Thu 5 Nov

Malaysia opens an AI-powered online centre for ASEAN startups

10,400startups on the platform behind the new ASEAN centre

Malaysia launched an online centre that uses AI to match startups with partners across ASEAN. Banks backed a big AI in Johor. A UNESCO study warns chatbots may be weaker in languages other than English.

Key points

Malaysia: at the Cradle LIVE! summit in Kuala Lumpur on 4 November, the science ministry MOSTI and Cradle launched the Startup ASEAN Digital Centre of Excellence. It adds seven tools, including AI matching, to a platform with more than 52,000 users and 10,400 startups.
Malaysia: Singapore's Princeton Digital Group got from six banks, including Maybank, Affin Bank, HSBC and UOB. The money helps finish its 200 JH1 campus at Sedenak Tech Park in Johor, built for AI and work.
Region: a UNESCO and University College London study of up to 33 AI models found many dropped correct climate facts when users pushed back, a habit called . Answers were also weaker in languages other than English. Full data comes out on 16 November, so treat it as an early warning.

Why it matters: Malaysia is building both the data centres and the startup links for AI. If you ask a chatbot in Malay or another local language, double-check the answer.

Sources (3)

Malaysia launches ASEAN startup digital centre at Cradle LIVE! Summit 2026 · TechNode Global
Singapore's Princeton Digital secures green financing for data hub in Johor · TechNode Global
AI chatbots cave on climate facts under pressure, and SEA's languages may fare worst · e27

Today

Saturday 7 Nov

Friday 6 Nov

Thursday 5 Nov