Singapore AI News & Daily Briefing

Singapore is updating its National AI Strategy to accelerate AI adoption across key sectors like healthcare, finance, and transport. The government is rolling out new initiatives to support businesses and workers. This means more AI tools in public services and potentially new job roles, but also a need for upskilling.

Applied Materials is building a mega facility in Singapore to strengthen the country's semiconductor manufacturing capabilities. This is a big deal because semiconductors are the brains behind everything from smartphones to AI systems. For Singaporeans, it means more high-tech jobs and a stronger position in the global chip supply chain, which affects the cost and availability of electronics.

Meta has launched a Business Agent that can recommend products and close sales directly inside WhatsApp, Messenger, and Instagram. It's now global, but Southeast Asia — where chat-based shopping is already huge — will be the real test. This could make shopping as easy as chatting with a friend, but also raises questions about privacy and spam.

A new report from JFrog finds that APAC companies, especially in Singapore, are using more software supply chain security and AI governance tools. This means businesses are getting better at spotting vulnerabilities in the software they buy or build. For the average user, this reduces the risk of data breaches from compromised apps.

AI workloads are driving up cloud storage costs unexpectedly due to data retrieval and transfer fees. Many businesses are moving some data back to on-premise storage to save money. Synology showcases local examples: I Love Taimei and LASALLE College in Singapore use on-premise setups to cut costs and keep data private. For businesses, it's a reminder to watch your cloud bill.

Google DeepMind and partners are putting up to $10 million in funding for researchers to study the safety of multiple AI agents interacting with each other. As more AI agents are deployed across different companies and systems, they could create unpredictable behaviors like economic disruptions or new security risks. This research aims to develop tools to monitor and control these systems before they become widespread, helping ensure the technology remains safe as it scales.

Companies like SpaceX, Google, and Starcloud are pushing to build AI data centers in orbit, but the physics of space makes it far harder and pricier than expected. The only way to cool chips in space is via radiation, requiring huge radiator surfaces, and radiation damages hardware. A cost analysis shows space-based computing is at least 10 times more expensive than on Earth, meaning this futuristic solution likely won't help with data center power shortages anytime soon.

OpenAI is acquiring Ona, a company that provides secure cloud execution environments, to let its Codex AI agent work persistently in the background even when your laptop is closed. This means Codex can handle long-running tasks like software modernization or security fixes over hours or days, giving organizations more control over their data and infrastructure.

BBVA, a global bank, has deployed OpenAI's ChatGPT Enterprise to over 100,000 employees, saving about 3 hours per person per week and achieving up to 80% efficiency gains in some tasks. The bank is now using AI to redesign everything from customer service to risk analysis, marking a major shift in how traditional banking operates. Employees have created over 20,000 custom assistants for specific workflows.

NVIDIA released Nemotron 3.5 Content Safety, a customizable AI model that checks both images and text for harmful content. It’s designed to help businesses deploy safer chatbots, image generators, and other AI tools by catching toxic, violent, or unsafe material. Unlike generic filters, this one can be tuned to a company’s specific rules.

OpenAI has formally backed the European Commission’s new Code of Practice on transparency for AI-generated content, which requires labeling and tracking of AI-made images and videos. The company uses a mix of metadata, watermarks, and verification tools to help people know if something was created by AI. This is part of the broader EU AI Act efforts to fight disinformation and protect election integrity.

A new whitepaper proposes a framework to measure the autonomy of wellness robots designed to assist seniors, based on seven dimensions of wellness and a six-level autonomy scale. This helps clarify what these robots can actually do independently — from helping with daily tasks to promoting social engagement. For families considering robot care for elderly relatives, this provides a clearer way to compare capabilities and trustworthiness.

Google DeepMind released Gemma 4 12B, a new AI model small enough to run on a regular laptop with 16GB of memory. Unlike most multimodal models, it processes images and audio directly without separate encoders, making it faster and more efficient. Developers can use it for agents and apps locally, and it's free to use under an open license.

Intel is planning to release processors in 2028 that integrate NVIDIA graphics technology directly onto the chip. This unusual partnership would combine Intel's CPU design with NVIDIA's integrated GPU (iGPU) technology, potentially offering better graphics performance in standard laptops and desktops without needing a separate graphics card.

Microsoft patched a critical vulnerability in its M365 Copilot AI platform that could have allowed hackers to steal two-factor authentication (2FA) codes and other sensitive data from emails. Security researchers discovered that attackers could send a malicious link to targets; when clicked, Copilot would secretly search the victim's emails and exfiltrate data including 2FA codes by embedding them in image requests that passed through Microsoft's Bing search service as a workaround. The root cause is that AI systems cannot distinguish between instructions from users and hidden commands in content they're processing, a fundamental flaw that guardrails cannot fully fix.

Singapore's Digital Development Minister Josephine Teo told an ASEAN summit that the region has strong foundations to accelerate AI adoption, but success depends on five pillars: infrastructure, workforce skills, societal participation, governance frameworks, and regional cooperation. She warned that overly restrictive data policies could stifle innovation, and called for ASEAN countries to coordinate rather than go it alone on AI development.

Singapore-based Wubble AI is taking a different path from AI music rivals like Suno, which was sued by major record labels for training on copyrighted music. Wubble pays musicians upfront to record original music specifically for training its model, hoping to avoid legal battles. Founded in June 2024, the startup sells royalty-free music to enterprises, helping companies skip the costly and time-consuming process of rights clearance. However, industry experts say "ethical AI music sourcing" is still loosely defined with no accepted standard.

Anthropic is expanding Project Glasswing to 150 additional partners across more than 15 countries, giving vetted organizations access to its Mythos Preview model to find software vulnerabilities. The program, launched with 50 partners in April after White House concerns about AI being used to discover flaws, has already uncovered over 10,000 high- or critical-severity bugs. The new group includes organizations in power, water, healthcare, and communications infrastructure. Anthropic says the AI tools can significantly shrink the time and cost of identifying software weaknesses.

Noam Shazeer, who co-led Google's Gemini AI project, is leaving to join OpenAI. Shazeer previously worked at Google on AI research before leaving and returning. This move intensifies competition between the two AI giants as both race to build more capable models.

South Korean gaming companies are diversifying their game portfolios and targeting global markets. Traditionally known for mobile massively multiplayer online role-playing games (MMORPGs), these developers are now exploring new genres and platforms. NCSoft plans to release seven in-house titles in the second half of 2025. The game, built on Unreal Engine 5, will feature action-oriented gameplay and improved aerial movement. NCSoft will also conduct closed beta testing for its upcoming shooter game, LLL, which is scheduled for release this year. Netmarble, another key player in the industry, recently showcased its cooperative shooter project Evilbane at IGN Live 2025 and will launch The Seven Deadly Sins: Origin across major platforms. Pearl Abyss and Kakao Games are pursuing global markets

Australian medical testing company Australian Clinical Labs confirmed that hackers stole limited data from around 280,000 people through a breach at an external IT provider used by its SunDoctors unit. The compromised data includes basic contact details and some health information related to skin cancer checks. No evidence suggests the data has been published online.

_Welcome to Tech in Asia’s daily newsletter, your essential dose of Asia’s tech and startup buzz. Not yet a subscriber? [Register here](https://www.techinasia.com/subscription?utm\_source=web&utm\_medium=article&utm\_campaign=dailynewsletter). Got a story tip? Send it to [email protected]._ ## In focus * ​​[Loans over users](https://www.techinasia.com/tonik-lowers-loan-growth-forecast-hits-cash-flow-breakeven-q1) for Tonik * ​​[Premium experiences](https://www.techinasia.com/plaud-hits-us100m-arr-singapore-key-hub) over cheap pricing for Plaud * GCash pushes through with [long-awaited IPO](https://www.techinasia.com/news/gcash-parent-mynt-gets-approval-to-pursue-ipo) * * * Hello reader, Walk around a mall in Manila and you’ll usually find a digital banking kiosk with agents convincing

Meta's AI product lead has departed as the company undergoes internal restructuring of its AI division. This marks another high-profile executive exit from a major tech firm, reflecting ongoing volatility in Big Tech's AI teams. Such departures can impact product roadmap consistency and signal deeper internal tensions around AI strategy.

Equinix has opened HK6, a US$124 million AI-ready data center in Hong Kong with liquid cooling technology designed for high-density AI workloads. The facility can scale to 3,550 cabinets and connects directly to the Hong Kong-Shenzhen Innovation and Technology Park to support cross-border AI development in the Greater Bay Area. Through Equinix's AI Discovery Hub with HPE and NVIDIA, enterprises can test and validate AI applications before committing to full deployment. The facility runs on 100% renewable energy.

OpenAI, Anthropic, and Google are working together through the Frontier Model Forum to detect and stop Chinese AI companies from copying their models. Anthropic says DeepSeek, Moonshot, and MiniMax used 24,000 fake accounts to extract 16 million responses from Claude to train rival systems. The companies say this "distillation" threatens their business models and raises national security concerns, as copied models without safety guardrails could power cyberattacks or disinformation. The rare collaboration walks a fine antitrust line but has support from the US government.

South Korean chemical giant LG Chem is partnering with a UK biotech firm to use AI for faster drug discovery. The collaboration aims to accelerate developing new medicines by using artificial intelligence to analyze data and identify potential compounds. This reflects a broader trend of major corporations adopting AI to speed up traditionally slow pharmaceutical research.

Japanese chip equipment makers like Advantest and Tokyo Electron are crushing earnings forecasts as AI drives unprecedented demand for memory chips and production tools. Advantest raised its full-year profit forecast by 25% to US$2.4 billion on strong sales of high-bandwidth memory testers. Meanwhile, car-chip suppliers like Renesas are struggling with weak auto demand. Japan has poured US$36.9 billion into reviving its chip industry since 2021, and 300mm fab equipment spending is projected to reach US$138 billion globally by 2028.

Singapore's IMDA has launched a $48 million four-year programme to help local media firms create Singapore-centric content and experiment with AI tools. The scheme will support short-form videos, micro-dramas, and AI-assisted production workflows. Media professionals can also use SkillsFuture credits to take 55 curated AI courses covering content creation, storytelling, and audio production. The first batch of 117 media firms received accreditation in June.

Datasaur, a data labeling startup, is pivoting to deploying private large language models for enterprises. After Meta's investment in Scale AI made some clients nervous about sharing proprietary data, Datasaur saw demand surge. The company now helps businesses run custom AI within their own environment instead of relying on shared public models—targeting firms that can't afford Big Tech's high consulting fees. Datasaur works with clients including AI Singapore and is targeting profitability by year's end.

TikTok is rolling out new AI-powered advertising tools specifically for the Southeast Asian market, aiming to help brands create short-form drama content more efficiently. The platform sees this as a way to capture a growing segment of viewers who prefer quick, engaging video stories.

Microsoft is preparing to unveil several in-house AI models at its Build developer conference, including a coding model for GitHub Copilot and models for transcription, reasoning, speech, and images. This extends Microsoft's existing Phi-4 model work and marks a push to reduce reliance on OpenAI, whose partnership has drawn US antitrust scrutiny.

Apple is set to raise prices on its devices as the AI boom drives up costs for advanced chips. Memory and processor prices have climbed due to massive demand from AI applications, and these higher component costs will likely show up in next-generation iPhones, MacBooks, and other Apple products. For everyday consumers, this could mean spending more when it's time to upgrade.

Tencent Cloud launched an upgraded AIoT 2.0 product suite that integrates voice and video AI into smart devices. The suite includes TWeTalk (AI voice interaction with speech-to-text, emotion recognition, WeChat integration) and TWeSee (video analytics for smart cameras). Partners like Alpha Group and makers of Talking Tom AI Buds are already using these tools for smart toys, robotics, and surveillance products. The global AIoT market is projected to reach $79 billion by 2030.

Singapore's Minister for Digital Development Josephine Teo warned Southeast Asian nations at the Asia Economic Summit that two major pitfalls could derail AI adoption: overly strict data rules that choke innovation, and a misguided push for full AI sovereignty that most countries can't afford. She stressed that governments must balance protecting citizens with enabling businesses — especially smaller ones — to use AI effectively.

Equinix and Cisco have partnered to launch a sovereign AI infrastructure hub in Singapore, integrating Cisco's Secure AI Factory powered by NVIDIA into Equinix's local data centers. This means businesses in Singapore can now run AI workloads on infrastructure that stays within the country, meeting local data residency rules without relying on overseas cloud services. For you, this could mean faster, more reliable AI-powered services from local companies—from banking apps to government tools—since the data never leaves Singapore's borders.

The US government has suddenly restricted access to Anthropic's latest Fable 5 AI models for European and Canadian users, essentially hitting a "kill switch" on international access. European and Canadian leaders are scrambling as their businesses and researchers lose access to these AI tools. The move marks a sharp escalation in AI nationalism, with the US treating advanced AI models like critical technology that must be protected domestically.

Singapore is becoming a key battleground for AI talent. OpenAI leads with 37 open positions in Singapore (up from 22 in May), followed by Plaud with 26 roles. Other recruiters include Motional, AlphaSense, ElevenLabs, and Mistral. The city-state has committed nearly US$780 million to AI research over four years and is building a new AI startup cluster to attract frontier AI labs.

Former President Trump announced that Apple will partner with Intel to manufacture chips in the United States. This would mark a significant shift in Apple's supply chain strategy, potentially reducing reliance on overseas chip production.

Shanghai's stock exchange has updated its listing rules to make it easier for AI companies to go public. This is significant for Chinese AI firms like robotics company Unitree, which recently passed regulatory review for a Shanghai IPO. The rule changes signal China's push to attract more AI sector listings amid global competition for tech listings.

CrowdStrike has acquired identity security firm SGNL for $740 million, expanding its AI-driven risk coverage capabilities into cloud environments. The move strengthens CrowdStrike's partnership with AWS by adding identity threat detection tools that help businesses protect against breaches involving compromised credentials.

OpenAI's research lead argues that companies misunderstand AI progress by relying on public benchmark scores rather than internal testing. She says models improve continuously in labs long before users notice, and that testing teams should measure actual workplace tasks, safety, and long-term autonomous work rather than chasing leaderboard rankings. Physical-world testing (like AI controlling lab robots to design proteins) creates a competitive advantage, as 'pain is the moat.' The piece also notes that traditional tests fail to measure AI agents that work independently for days or weeks.

Singapore's government has launched an S$48 million programme to help local media companies adopt AI tools and create digital content. Run by IMDA and MDDI, the initiative funds experimentation with formats like micro dramas, AI-powered animation and visual effects, and offers up to 90% course fee subsidies for workers aged 40 and above across 55 AI training courses. Early trials produced over 2,000 short videos generating 200 million local views.

Meta has signed deals to get AI computing power from Crusoe, a data centre developer, at two US locations in Texas and Missouri. The combined capacity will be roughly 1.6 gigawatts - enough to power about 750,000 US homes. This is part of Meta's US$600 billion infrastructure investment pledge over three years to support its AI ambitions.

The White House flagged concerns that Anthropic's latest AI was 'jailbroken' — meaning someone bypassed its safety guardrails to make it do things it normally wouldn't. In plain terms, jailbreaking is like unlocking your phone to remove restrictions, but for AI: hackers find loopholes to make the system ignore rules meant to prevent harmful outputs. This matters because it shows even top AI labs struggle to keep safety measures intact, and government scrutiny on AI safety is heating up.

Apple is adding a feature to its Passwords app that will automatically detect and update weak or compromised passwords for you using AI. Instead of manually changing passwords one by one, users tap a 'Fix Passwords' button and Apple handles it in the background. Security experts are cautiously optimistic but want to see how it performs in practice, since AI-generated passwords aren't always as strong as expected.

OpenAI rolled out new analytics and spending controls for its business platform, ChatGPT Enterprise. Companies can now see exactly who's using AI, set spending limits per team or employee, and spot usage trends before costs spiral. As more offices adopt AI tools, bosses need better ways to track budgets - and these new dashboard features give them that control.

VivaTech 2026 in Paris is showcasing the latest in AI-powered robotics and consumer tech. Unitree's humanoid robots demonstrated boxing and balance skills, while Samsung displayed smart fridges that track groceries and auto-generate shopping lists. The three-day event draws startups and industry leaders focused on real-world AI applications across logistics, manufacturing, and daily life.

Commonwealth Bank of Australia has hired new technology leaders to drive its AI and tech strategy. Banks worldwide are racing to adopt AI for customer service, fraud detection, and operations, so these executive moves signal where the industry is heading—and could eventually affect how Australians bank.

The Trump administration has issued an export control directive forcing Anthropic to take its most advanced Claude models (Mythos and Fable 5) offline, and nearly a week later the two sides still cannot agree on how to restore them. The White House claims it found a way to "jailbreak" Fable 5, but the specifics remain unclear. This raises concerns about how US AI regulations are being made without clear rules or consistent enforcement - creating uncertainty for the entire AI industry.

OpenAI is hiring two high-profile figures as it gears up for its IPO: Noam Shazeer, the AI legend who co-authored the seminal Transformer paper and just left Google DeepMind, and Dean Ball, a former White House AI policy official. They are creating a new 'Strategic Futures' team to handle AI governance, catastrophic risk, and the company's relationship with governments. This comes as rival Anthropic faces US government export restrictions on its latest models.

Amazon Web Services is in talks to sell its homegrown Trainium AI chips to other companies, potentially creating a $50 billion business that would directly challenge Nvidia's dominance in the AI chip market. AWS CEO Andy Jassy said demand for its chips is so high that current and next-generation capacity sold out almost instantly. While Nvidia is currently on a $326 billion revenue run rate, an Amazon entry into chip sales would mark one of the biggest competitive threats Nvidia has faced.

Google DeepMind treats its own AI agents as potential insider threats. The company's new 'AI Control Roadmap' links security measures to AI capability levels, and an analysis of one million coding tasks found most problems come from overzealous agents, not malicious intent. DeepMind warns the window to establish global AI security standards is closing quickly.

Hackers broke into a Texas state government vendor and stole the personal data of 3 million people—including driver's license numbers, passport details, addresses, phone numbers, and email. The breach happened through the Texas Parks & Wildlife Department's license sales system. This is one of the biggest government data breaches this year, and those affected face increased risk of identity theft and phishing scams.

SK Telecom had access to Anthropic's AI model Claude through a partnership program until the White House intervened. US officials flagged concerns about ties between the South Korean telecom giant and China, leading Anthropic to restrict access. The incident shows how geopolitics is now shaping which companies can work with leading AI labs.

Senator Bernie Sanders has introduced a US bill that would give Americans $1,000 annually from an AI sovereign wealth fund, arguing AI is a public resource. The fund would be funded by taking a 50% stake in top AI labs.

Apple has fixed a serious security flaw in its Beats Studio Buds earbuds that could have let nearby hackers secretly listen through your phone's microphone. The bug affected Bluetooth authentication in chips made by Airoha Systems and scored 8.8 out of 10 on severity. Apple pushed out a firmware update automatically, but if you use wireless earbuds, it's worth checking your settings to confirm you're running the latest version.

 Google/Amazon/ZDNET _\_Follow ZDNET:\_[Add us as a preferred source](https://cc.zdnet.com/v1/otc/00hQi47eqnEWQ6T9d4QLBUc?element=BODY&element\_label=Add+us+as+a+preferred+source&module=LINK&object\_type=text-link&object\_uuid=e188be20-525c-40c5-a567-83fa72410cb6&position=1&template=article&track\_code=\_\_COM\_CLICK\_ID\_\_&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view\_instance\_uuid=04408701-6fd6-48dc-a316-4dfcd21e921a&object\_version=c5206502-4e93-44ee-a656-b56f00458ae0)\_on Google.\__ * * * ### ZDNET's key takeaways * New Google Home Speaker arrives Ju

US federal regulators ordered grid operators to fast-track connections for data centers and other large electricity users. The Federal Energy Regulatory Commission directed six major operators to ensure data centers can connect 'in a timely and orderly manner,' with data centers covering the costs. The move aims to address severe backlogs — at one point, grid connection requests exceeded the entire existing power plant capacity. Meanwhile, electricity prices have surged up to 267% compared to five years ago.

AI company Baseten is reportedly raising $1.5 billion at a $13 billion valuation, just five months after a $300 million round. The startup helps businesses run AI models quickly and cheaply by routing requests to the best-fit model, including cheaper open-source options. The funding shows investors remain hungry to back companies that make AI cheaper and easier to deploy.

YouTube is cracking down on AI-generated 'slop' by tweaking its algorithm to favour videos with real human faces. This is hurting legitimate 'faceless' creators—people who don't appear on camera but produce genuine content. The platform is also testing a pop-up that asks viewers to rate if a video feels like AI slop, but critics say crowdsourcing detection is flawed and might even train better AI fakes. For everyday users, this means fewer recommendations for faceless channels, even if they're human-made, reducing variety on the platform.

Fed up with constant screen time and addictive algorithms, more people are embracing 'slow tech' — ditching smartphones for flip phones, iPods, and screenless wearables. They want friction that creates boundaries, not seamless engagement. Even AI tools are being used to help people disconnect. This movement is a direct backlash against the attention economy built by the very people who now regret it.

Snap is spinning off its AI video team into a new company called Dotmo, which will focus on AI-powered interactive gaming experiences. Snap cited the high costs of developing AI internally as a key reason for the move. Snap's CTO Bobby Murphy is personally investing as lead investor, while Snap retains a significant equity stake and has licensed its technology to Dotmo. The Dotmo team consists of current Snap staff leaving to launch the venture.

A new Reuters Institute report shows 10% of people worldwide now use AI chatbots for news weekly, up from 7% last year. However, only 4% actually click through to read the original source. The low trust in AI-generated news summaries could reshape how media outlets reach audiences — if publishers can't get readers past the chatbot, they lose ad revenue and influence.

# I'm a smart home reviewer, and these are the only deals I'm shopping this Prime Day | ZDNET 03 days 15 hours 27 minutes 08 seconds X Trending * [I wore the world's first HDR10 smart glasses](https://www.zdnet.com/article/rayneo-air-4-pro-hdr10-smart-glasses-ces/) * [TCL's new E Ink tablet beats the Remarkable and Kindle](https://www.zdnet.com/article/tcl-note-a1-nxtpaper-ces/) * [Anker's new charger is one of the most unique I've ever seen](https://www.zdnet.com/article/anker-wireless-chargers-ces-2026/) * [Best laptop cooling pads](https://www.zdnet.com/home-and-office/best-laptop-cooling-pad/) * [Best flip phones](https://www.zdnet.com/article/best-flip-phone/) * [Roku TV vs Fire Stick](https://www.zdnet.com/home-and-office/home-entertainment/roku-tv-vs-fire-stick/) * [Galaxy Buds 3 Pr

Vietnamese software firm Luvina Software has opened its Asia-Pacific headquarters in Singapore, betting on the city-state as a gateway to serve enterprise clients across the region. Founded in Hanoi in 2004, Luvina brings over 750 engineers and has already landed more than a dozen Singapore projects. The firm combines Singapore's governance credibility with Vietnam's cost-efficient engineering talent—a model aimed at helping businesses modernize legacy systems and adopt AI without the premium price tag of Western IT consultancies.

A new ADP workforce report shows that contrary to popular belief, Singapore's youngest workers aged 18-26 are actually the LEAST frequent AI users at just 20% daily usage, behind those in their 30s (25%) and 40s-50s (23%). Despite this, more Singapore workers feel confident about AI's job impact than before—only 10% remain uncertain, down from 19% last year. The data also reveals a paradox: daily AI users report feeling less productive even as they're more engaged, highlighting the challenge of adjusting to AI-assisted work.

Southeast Asia is well-positioned to accelerate artificial intelligence (AI) adoption thanks to growing digital infrastructure, a young tech-savvy population and increasing regional cooperation, Singapore’s Digital Development and Information Minister Josephine Teo said on Thursday. Speaking at a digital summit in Jakarta, Teo said ASEAN was making progress across the key pillars needed for widespread AI deployment, including infrastructure, workforce capabilities, governance frameworks and international cooperation. “Overall, Southeast Asia is in a good position to deploy AI widely,” she said. “But we will not get there automatically.” Teo said data center capacity across the region is expected to more than triple between 2025 and 2030, while expanding submarine cable networks are strengt

Nvidia's CEO Jensen Huang announced that Samsung, SK hynix, and Micron have all qualified to supply HBM4 memory chips for Nvidia's upcoming Vera Rubin AI platform, set for delivery in Q3. This is a big deal because HBM4 doubles data transfer speeds compared to current HBM3E chips, making AI systems faster and more powerful. With all three major memory makers now in the game, expect more competition—and potentially more stable pricing—for the high-performance memory that powers everything from ChatGPT-style chatbots to enterprise AI tools.

Anthropic is expanding access to Claude Fable 5, its newest AI model, to enterprise customers and paid users after initially restricting it over safety concerns. The company added new safeguards to block risky responses in areas like cybersecurity and biology. Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens. The rollout comes as Anthropic has confidentially filed for an IPO.

China's government has announced new measures to encourage deeper integration of AI technology into consumer markets, though the specific policy details were not available in this report. The initiative appears aimed at boosting AI adoption across retail, services, and daily life applications in the world's largest consumer market.

OpenAI's reasoning model helped doctors find genetic diagnoses for 18 children out of 376 previously unsolved rare disease cases. The AI didn't diagnose anyone—it suggested leads that human experts then confirmed through clinical testing. This research, published in NEJM AI, shows AI can act as a tireless research assistant that links a patient's symptoms, genetic data, and new scientific findings to help solve medical mysteries that had stumped specialists for years. For families of children with undiagnosed rare diseases, this approach could eventually become a new tool for finding answers.

Enterprises need to clearly assign ownership and accountability for AI agents before letting them act across business systems, according to Matthew Hardman, APAC CTO at Hitachi Vantara. He warns that fragmented data makes AI governance difficult, and accountability should not rest solely with the AI model itself. Companies must establish who is responsible when AI agents make decisions or errors.

Coinbase CEO Brian Armstrong is pushing crypto as the payment infrastructure for AI agents — software programs that autonomously perform economic tasks. The idea is that AI agents need to pay for compute, hire other agents, and monetize services, but traditional banking systems can't handle the tiny, frequent transactions this requires. Coinbase is building tools like USDC stablecoin and its Base network to enable machine-to-machine payments, with open standards like x402 allowing AI to autonomously pay for API access and services.

ChatGPT now builds a detailed profile of you by analyzing your entire chat history, not just saved facts. The feature, called 'Dreaming,' synthesizes your preferences and habits into summaries that guide future answers. The catch: the AI sometimes gets things wrong, remembering details you never shared or creating false context. Users can review and delete memories, but even turning off the feature doesn't erase everything.

Singapore's central bank chief has warned that global markets are becoming too reliant on AI investments, creating potential vulnerabilities if expectations are reassessed. Speaking at the Lujiazui Forum in Shanghai, MAS managing director Chia Der Jiun said returns on AI spending remain uncertain while costs for energy and chips keep rising. He urged economies to build resilience and deepen regional cooperation as governments have less fiscal room to respond to future shocks.

Deutsche Bank says AI has cut its internal tech project timelines from years down to just months. The bank uses AI tools to clear backlogs and automate data analysis, though it closely monitors token usage costs as AI firms shift to usage-based pricing. Engineers get token quotas and must prove value for extra capacity — a discipline similar to how companies learned to manage cloud spending.

AI is learning to understand and generate speech better than ever, but the quality depends entirely on the data used to train it. Researchers are developing ways to tell which speech datasets will produce useful AI versus ones full of errors or bias. For everyday users, this means voice assistants and transcription tools will only be as good as the data builders feed them.

Vietnam International Commercial Joint Stock Bank (VIB) is deploying Salesforce to unify customer data across its 7 million customers and 200 branches, building toward AI-assisted banking workflows. The platform will enable real-time 360-degree customer views, faster digital card approvals, and personalised marketing. VIB is exploring AI-assisted customer segmentation and banker recommendations with human oversight to meet Vietnam's banking compliance requirements.

Intel shares jumped 9% after Trump announced Apple agreed to design and build its chips in the US using Intel's facilities. Since returning to office, Trump has pushed tech companies to move chip production out of Asia and into America. Nvidia also agreed to build its first-level chips with Intel, and Elon Musk's TerraFab—the world's largest chip factory—will use Intel's technology team. Intel's share price has surged over 400% since the US government invested $8.9 billion in the company last August.

Joshua Baer, founder of Capital Factory, died in a plane crash. He built Austin into a national AI hub by connecting startups to the U.S. Army, corporations like IBM and AWS, and major events like SXSW. His portfolio includes over 100 AI companies spanning robotics, healthcare AI, defense AI, and industrial automation. While he never shipped a foundation model, his legacy is the human infrastructure—mentorship, introductions, early funding—that let ambitious AI companies emerge. The article argues innovation often starts with a person who takes a call, not just a better model.

Architect Labs raised $24 million in seed funding to use AI to speed up and cheapen custom chip design. Currently, designing custom chips takes about two years and costs hundreds of millions. The company targets cloud giants like Amazon and Google who want alternatives to Nvidia hardware. Investors include Google DeepMind's Jeff Dean, OpenAI, and Nvidia executives.

AI is making it easier for companies to replace full-time employees with contractors, gig workers, and freelancers — workers with fewer protections and benefits. The article shows how Klarna uses AI chatbots for basic queries while gig workers handle complex ones, how musicians train AI that replaces them, and how nurses are being pushed onto "Uber for nursing" apps with lower pay and fewer protections. Labor experts warn this "gigification" of work could spread to white-collar jobs.

Google, Microsoft, and other tech giants are联手 creating ARD (Agentic Resource Discovery), essentially a search engine for AI agents. Think of it like an app store, but instead of humans downloading apps, AI agents can automatically discover and connect to other AI tools, skills, and services across the internet. The system works by having companies publish catalogs of their AI capabilities, which registries then index so other agents can find them—similar to how DNS works for websites. GitHub and Hugging Face already launched tools built on ARD. However, security experts warn this could create new targets for hackers, since compromising a catalog could trick AI agents into using malicious tools.

Scientists built a chip that uses sound waves to mimic how the brain works—it could make AI hardware run faster and use far less power. Researchers at the University of Arizona created an "acoustic synapse" using aluminum rods and sound waves that achieved 96.7% accuracy on flower classification while using just one-tenth the energy of conventional neuromorphic chips. The device can also simulate brain chemicals like dopamine, potentially letting one circuit handle multiple tasks.

Adobe is rolling out AI agents across its Creative Cloud suite including Photoshop, Premiere, and other apps. Users can now describe what they want in plain language, and the AI handles complex multi-step tasks automatically. The agents also integrate with external AI platforms like ChatGPT and Claude.

The US government has imposed export controls on Anthropic's powerful new AI models, Fable 5 and Mythos 5, after Amazon researchers flagged a potential security vulnerability. Anthropic, which had warned that Mythos could act as a 'cyber-weapon,' took both models offline rather than risk foreign nationals accessing them. The company was given just 90 minutes to comply. The move shows the US is now treating frontier AI as a national security concern, though experts say the hasty rollout is embarrassing — with one cybersecurity leader noting 'Beijing is laughing at us.'

Intel stock jumped 10% after reports that Apple will partner with Intel to manufacture processors — marking a potential reversal after Apple abandoned Intel for its own chips years ago. Google and Nvidia are also reportedly exploring Intel for chip fabrication, riding a wave of demand as AI companies need more manufacturing options beyond TSMC. Intel's stock is up 250% this year as its turnaround gains traction.

Intel's stock jumped 10% after Trump announced Apple will work with Intel to manufacture chips, potentially for less critical components. Apple previously dropped Intel for its own custom chips, so this would mark a notable U-turn. Intel is also reportedly working with Google and exploring deals with Nvidia. The chipmaker's turnaround is being driven by massive AI demand and other chip manufacturers being unable to keep up with orders. Intel's stock is up over 250% this year alone.

Apple CEO Tim Cook has warned that integrating AI features into iPhones, Macs, and other Apple products will inevitably drive up prices for consumers. Cook said the company is doing its best to manage costs but acknowledged the situation has become unsustainable as AI hardware and development expenses rise.

Google just launched a $99 Home Speaker with its Gemini AI assistant built in, going head-to-head with Amazon's Echo Dot Max (currently on sale for $65 with Alexa+). Google's speaker may be smarter at answering complex questions and understanding corrections mid-sentence, while Amazon's may sound better for music. Both work as smart home hubs, but Google's also acts as a Thread border router for better device connectivity.

Two peer-reviewed studies in Nature show specialized AI systems matching or beating doctors at diagnosing diseases and recommending treatments in simulated patient cases. One catch: both systems run on outdated base models, raising questions about long-term reliability as technology advances. The findings suggest AI is clinically viable today but may need frequent retraining to stay sharp.

Vietnam's tech industry is moving beyond its traditional role as a low-cost outsourcing hub for foreign companies. Local developers are now building their own AI products and startups, shifting from executing other people's ideas to creating their own. This could mean higher-paying local tech jobs and new Vietnamese tech companies competing globally, marking a bigger shift in Southeast Asia's technology story.

U.S. energy regulators just approved a plan to connect AI data centers to the power grid much faster. Big tech companies like Google, Microsoft, Meta, and Amazon will cover the costs of any grid upgrades themselves. Data centers already consume about 5% of America's electricity and could triple that by 2035. The move aims to help the U.S. compete with China in AI, but critics warn it could push up electricity bills for regular households and strain local power supplies.

General Intuition, a New York AI startup teaching robots and AI agents to understand space and time, is in talks to raise $300 million at a $2 billion valuation. Backed by Jeff Bezos and Eric Schmidt, the company uses video game footage from Medal's platform to train AI on real-world movement. It's competing against heavyweights like Runway and Google's world model teams.

Microsoft is growing its AI business in China by selling access to OpenAI's models through its Azure cloud platform. This lets Chinese companies use advanced AI tools even though OpenAI's services are officially restricted in China. The arrangement raises geopolitical questions around whether American AI technology should flow into China through commercial channels, especially as U.S. policy tightens restrictions on direct AI exports.

US federal regulators ordered grid operators to fast-track connections for data centers and other large electricity users. The Federal Energy Regulatory Commission directed six major operators to ensure data centers can connect 'in a timely and orderly manner,' with data centers covering the costs. The move aims to address severe backlogs — at one point, grid connection requests exceeded the entire existing power plant capacity. Meanwhile, electricity prices have surged up to 267% compared to five years ago.

The White House flagged concerns that Anthropic's latest AI was 'jailbroken' — meaning someone bypassed its safety guardrails to make it do things it normally wouldn't. In plain terms, jailbreaking is like unlocking your phone to remove restrictions, but for AI: hackers find loopholes to make the system ignore rules meant to prevent harmful outputs. This matters because it shows even top AI labs struggle to keep safety measures intact, and government scrutiny on AI safety is heating up.

Amazon Web Services is in talks to sell its homegrown Trainium AI chips to other companies, potentially creating a $50 billion business that would directly challenge Nvidia's dominance in the AI chip market. AWS CEO Andy Jassy said demand for its chips is so high that current and next-generation capacity sold out almost instantly. While Nvidia is currently on a $326 billion revenue run rate, an Amazon entry into chip sales would mark one of the biggest competitive threats Nvidia has faced.

Google DeepMind treats its own AI agents as potential insider threats. The company's new 'AI Control Roadmap' links security measures to AI capability levels, and an analysis of one million coding tasks found most problems come from overzealous agents, not malicious intent. DeepMind warns the window to establish global AI security standards is closing quickly.

SK Telecom had access to Anthropic's AI model Claude through a partnership program until the White House intervened. US officials flagged concerns about ties between the South Korean telecom giant and China, leading Anthropic to restrict access. The incident shows how geopolitics is now shaping which companies can work with leading AI labs.

OpenAI rolled out new analytics and spending controls for its business platform, ChatGPT Enterprise. Companies can now see exactly who's using AI, set spending limits per team or employee, and spot usage trends before costs spiral. As more offices adopt AI tools, bosses need better ways to track budgets - and these new dashboard features give them that control.

OpenAI is hiring two high-profile figures as it gears up for its IPO: Noam Shazeer, the AI legend who co-authored the seminal Transformer paper and just left Google DeepMind, and Dean Ball, a former White House AI policy official. They are creating a new 'Strategic Futures' team to handle AI governance, catastrophic risk, and the company's relationship with governments. This comes as rival Anthropic faces US government export restrictions on its latest models.

Apple is adding a feature to its Passwords app that will automatically detect and update weak or compromised passwords for you using AI. Instead of manually changing passwords one by one, users tap a 'Fix Passwords' button and Apple handles it in the background. Security experts are cautiously optimistic but want to see how it performs in practice, since AI-generated passwords aren't always as strong as expected.

Snap is spinning off its AI video team into a new company called Dotmo, which will focus on AI-powered interactive gaming experiences. Snap cited the high costs of developing AI internally as a key reason for the move. Snap's CTO Bobby Murphy is personally investing as lead investor, while Snap retains a significant equity stake and has licensed its technology to Dotmo. The Dotmo team consists of current Snap staff leaving to launch the venture.