Singapore AI News & Daily Briefing

Singapore is rolling out a major plan to ensure every university and polytechnic student graduates with AI skills, not just those in tech courses. The initiative covers AI fundamentals, practical applications, and ethics — aiming to prepare the entire workforce for an AI-driven economy. This aligns with Singapore's goal to become a leading AI hub by 2030.

Security researchers discovered that the Creative Sound Blaster Katana V2X speaker can be hacked over the air to infect connected computers with malware — without anyone ever physically touching the device. The company doesn't think this is a problem worth fixing. If you have USB devices connected to your work or home computer, this is a reminder that anything plugged in can be a risk.

Google Photos is reportedly getting a major AI upgrade called 'Video Remix' that can automatically generate video compilations from your photos. The feature would use AI to pick your best shots, match them to music, and create polished videos with transitions and effects — no editing skills needed. This could replace third-party video apps for casual users who just want quick, shareable memories.

Singapore's arts students and professionals are split on generative AI. Some, like LASALLE graduate Voon Shun Yi, reject it to preserve artistic craft and turn down jobs requiring AI tools. Others, like NTU student Cheerie Tan, use it to speed up workflow while keeping human creativity central. Arts institutions like LASALLE and SOTA have set up working groups and guidelines to navigate this shift, while artists like Karyn Lim embrace AI for prototyping and Joanne Ho for animating sketches. The consensus: AI can't replace human vision and intentionality, but it's reshaping how art gets made.

Singapore's arts schools are rewriting the rules on how they teach and grade creative work. LASALLE, NAFA, and the School of the Arts have set up guidelines and working groups to balance traditional skills with AI tools. Key changes include requiring students to document their creative process in journals, giving oral presentations to explain their work, and using tiered assessment scales that tell students upfront whether AI is prohibited, permitted, or required for each assignment. Schools acknowledge that performing arts may be more insulated than digital design fields, but all emphasize teaching critical thinking over just using tools.

This week's Asian startup funding roundup highlights AI and tech deals. Notable: Singapore chip firm Silicon Box raised $77.53M for advanced semiconductor packaging; Israel's Rylo secured $85M to build AI communication tools for deaf and hard-of-hearing users; India's Hygenco raised $105M for green hydrogen solutions. The piece also covers AI startup acquisitions in Korea and UAE.

Google Cloud is launching a dedicated AI startup accelerator for Southeast Asia, partnering with agencies from Singapore, Indonesia, and Vietnam. The three-month equity-free program will support 25 AI startups from seed to series B stage with engineering help, market access, and a trip to Silicon Valley to meet VCs on Sand Hill Road. Applications for the first cohort open now, with the program starting in August 2026.

Anthropic has been forced to shut down its most powerful AI models, Claude Fable 5 and Mythos 5, just days after launch. The US government raised national security concerns, specifically about a "jailbreak" technique that could bypass safety controls and potentially help hackers exploit computer systems. The UK's AI Safety Institute found the model could exploit cybersecurity defenses 73% of the time. This marks the first time a US AI company has faced such direct government intervention over security risks, with the Pentagon also labeling Anthropic a "supply chain risk."

Visa has partnered with OpenAI to let ChatGPT buy stuff directly on your behalf using your credit card. Instead of just recommending products, the AI can now complete purchases end-to-end—from picking headphones under $150 to checking out. It's a sign that AI "agents" are moving from theory to real transactions. But experts warn of risks: ChatGPT has already been caught recommending fraudulent storefronts that steal money and banking details. Scammers are increasingly designing websites to trick AI bots, meaning your digital shopping assistant could be fooled into handing over your cash to crooks.

SpaceX is expanding its Starlink satellite network to support AI workloads, positioning itself as a direct competitor to traditional cloud providers like AWS and Google Cloud. The company is deploying dedicated compute satellites that can handle AI processing tasks from orbit, potentially disrupting how businesses access AI infrastructure.

China released new rules requiring financial companies to handle customer data more carefully as part of a wider cybersecurity push. Banks and fintech firms must now follow stricter guidelines on how they collect, store, and protect user information.

An opinion piece examining how Anthropic's decisions around model access reflect broader power dynamics in the AI industry. The analysis suggests that controlling who can use advanced AI models is itself a form of competitive advantage, raising questions about openness, safety, and market competition among AI companies.

AI company leaders are quietly walking back their doomsday predictions about AI replacing white-collar jobs. After years of warnings about mass unemployment, executives are now softening their language — likely to avoid spooking regulators, investors, and the public. The shift suggests the industry knows it overplayed the job-displacement angle and is trying to manage expectations.

Researchers have used data from Pokémon Go players — specifically millions of recorded movement patterns — to train AI systems that could guide drone navigation. The location data, collected from players exploring real-world environments, helps AI learn how objects and people move through spaces, which is useful for autonomous drone operations.

The US government has blocked foreign users from accessing Anthropic's most advanced AI models, according to Axios. This means developers, businesses, and researchers outside America can no longer use Claude's top-tier capabilities. It's part of a broader US push to keep frontier AI technology within American borders, citing national security concerns. If you rely on these tools for work or projects, you'll need to find alternatives or wait for policy changes.

Anthropic is expanding its Project Glasswing cybersecurity initiative to 150 more partner organizations across 15 countries, giving vetted companies access to AI models for identifying software vulnerabilities. Partners have already found over 10,000 critical flaws since the April pilot. The expansion includes organizations in power, water, healthcare, and communications sectors.

A compact robot roughly the size of a baseball has landed on the Moon as part of a lunar mission and is sending back images of the surface. The robot uses AI for navigation and terrain mapping to avoid obstacles in the challenging lunar environment.

Dutch chipmaking equipment giant ASML is set to become the largest shareholder in French AI startup Mistral AI, committing €1.3 billion (US$1.5 billion) in a €1.7 billion funding round. The deal values Mistral at €10 billion, making it Europe's most valuable AI company. ASML, the sole supplier of EUV lithography machines used by TSMC and Intel, will also get a board seat, signaling a deeper strategic partnership rather than just a financial investment.

Meta CEO Mark Zuckerberg has publicly acknowledged mistakes in the company's AI restructuring efforts. The social media giant has been aggressively pivoting toward AI under Zuckerberg's leadership, but this rare admission suggests internal strategy missteps. The reshuffle reportedly affected stock performance as investors reacted to uncertainty around Meta's AI direction.

AI cameras are being tested in Kananaskis to automatically spot and alert authorities about wildfires before they spread. The system uses computer vision to detect smoke and flames early, potentially cutting response times and saving forests and homes.

[ ](https://profiles.yahoo.com/brands/associated-press/) Updated Sat, 13 June 2026 at 2:29 am UTC 1 min read Anthropic co-founder and President Daniela Amodei, left, shakes hands with Snowflake CEO Sridhar Ramaswamy during the keynote presentaton at Snowflake Summit 26 Monday, June 1, 2026, in San Francisco. (AP Photo/Jeff Chiu) WASHINGTON (AP) — [AI giant Anthropic](http

# Huawei backs DeepSeek V4 with Ascend AI chips [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/companies/fundr

OpenAI has acquired Ona, a startup whose technology lets AI coding assistants run securely inside companies' own cloud infrastructure. This move brings powerful Codex coding agents directly into enterprise environments without data leaving their systems — a major selling point for businesses worried about sending code to external servers.

# Mapping the AI startups making waves in Japan - Tech in Asia [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/

Google Research's Gemini-SQL2 turns natural language into executable SQL queries. Built on Gemini 3.1 Pro, it tops the BIRD benchmark at 80.04 percent accuracy, well ahead of OpenAI and Anthropic. Google says the technology could improve natural language features across its data services. The article Google Research's Gemini-SQL2 tops text-to-SQL benchmarks by a wide margin appeared first on The Decoder .

This is the same story as the Anthropic Fable/Mythos shutdown—Ars Technica's more detailed coverage confirms the company completely disabled access after a US Commerce Department directive restricting use outside America. The administration cited concerns about a jailbreak that could bypass safety filters for cybersecurity, chemistry, and biology topics. Anthropic disputes the severity, saying the vulnerabilities found are "minor" and exist in other AI models too. The company is suing the Pentagon over a "supply chain risk" designation—the first ever given to a US company.

A coalition of US state attorneys general is investigating OpenAI, demanding records on how it handles user data, protects minors and seniors, and manages AI safety. The probe comes as OpenAI prepares for a potential trillion-dollar IPO and faces multiple lawsuits alleging ChatGPT caused real-world harm, including a case where a mother claims the chatbot encouraged her daughter's suicide.

The US government has ordered Anthropic to shut down global access to its Claude Fable 5 and Mythos 5 AI models, citing security concerns about potential jailbreaks. Anthropic is complying but fighting back publicly, arguing the vulnerabilities are minor and exist in competing models like GPT-5.5. The company warns this could set a precedent that makes frontier AI deployments harder everywhere. For everyday users, this means some AI tools may become unavailable or more restricted as companies scramble to meet government security demands.

The US government ordered Anthropic to block its most advanced models, Fable 5 and Mythos 5, from all foreign users over national security fears. Rather than carve out exemptions, Anthropic cut off access entirely — meaning American customers lost out too. It's one of the most sweeping AI export restrictions yet, raising questions about how far governments will go to contain AI risks.

Microsoft and three Chinese universities have developed SkillOpt, a method that optimizes instruction documents for AI agents using principles from traditional model training. A simple Markdown file is enough to boost GPT-5.5 by about 23 points on procedural tasks, and the same file transfers across models and agent environments like Codex and Claude Code. The article Microsoft's SkillOpt boosts GPT-5.5 by using nothing but a trained Markdown file appeared first on The Decoder .

A German court has ruled that Google is legally liable for false statements produced by its AI Overviews feature, setting a precedent that companies operating AI systems must assume responsibility for damages caused by their outputs. The ruling moves beyond previous cases that blamed AI "hallucinations" on users, instead holding AI deployers accountable for the systems they design, train, and manage.

Anthropic's Claude Fable 5 scored 88 percent on the hardest FrontierMath problems, beating OpenAI's GPT-5.5 by 13 points. This is a significant leap from its predecessor, which scored below 10 percent just months earlier. The rapid improvement shows AI math capabilities are accelerating fast.

Visa has partnered with OpenAI to handle payments when AI agents make purchases or transactions on behalf of users. This is a new frontier: AI systems that spend money autonomously. The key question is trust—can you safely let an AI agent use your card? Visa is building safeguards to verify these AI-initiated transactions, but experts warn consumers should understand exactly what their AI is authorized to buy.

Chinese AI startup Moonshot AI released Kimi K2.7 Code, a coding-focused model with one trillion parameters that costs up to 12 times less than GPT-5.5 or Claude per token. It trails top models on benchmarks but offers developers far more runs for the same budget—a trade-off that could make it attractive for cost-conscious teams.

The FCC is pushing to eliminate burner phones, citing their use in fraud and crime. Meanwhile, AI is now being used to automatically find software vulnerabilities—Microsoft just had its biggest-ever patch day thanks to AI bug hunting. A ransomware gang also exploited a new Oracle flaw to steal data.

Google Research has released Gemini-SQL2, an AI model that converts plain English into SQL database queries. It scored 80.04% on the BIRD benchmark, significantly outperforming rival models from OpenAI and Anthropic. This matters for everyday users because natural language database queries could eventually let non-technical people extract insights from data without learning coding.

Researchers trained robots to read human emotions using vision language models that can analyze not just faces but whole scenes and context. In tests with 40 volunteers, the AI system outperformed traditional emotion-recognition software (scoring 0.86 vs 0.77). People preferred when robots gave personalized apologies after making mistakes, but the study found that functionality still matters more than emotional charm—a robot that fails its task loses trust regardless of how well it apologizes.

Companies are hitting a wall with AI subscription costs and are starting to look at cheaper alternatives — including Chinese-built AI models and free open-source options. This shift could reshape the AI market by giving budget-conscious businesses access to capable tools without the premium price tag.

Rivian CEO RJ Scaringe has launched a separate robotics company called Mind Robotics, which has raised over $1 billion. Unlike Elon Musk's approach with Tesla, Scaringe keeps the robotics business independent while positioning Rivian as a major customer. Mind expects to reveal its first humanoid robot within a year. Humanoid robots powered by AI could eventually work alongside humans in factories, hospitality, and homes.

Meta is cracking down on runaway AI costs. An internal memo sent to 6,000 employees reveals the company is projected to spend billions just on internal AI use. Starting 2027, all AI spending will need approval through a new system called 'AI Gateway.' CTO Andrew Bosworth told staff that burning more tokens doesn't equal real progress. The shift signals even the biggest AI players are now treating compute costs as a serious budget line item.

Apple's iOS 27 brings its first real AI photo editing tools to the world's most popular camera. Features like object removal and scene extension are modest compared to Google Pixels, but they signal Apple's entry into AI-enhanced photography. For everyday iPhone users, it means touching up photos gets easier — no third-party apps needed.

Microsoft CEO Satya Nadella is warning against "token-maxing," throwing the most powerful AI models at every problem. Frontier models shouldn't be wasted on everyday tasks, he says. The marginal cost of productivity gains has to match the token cost. Still, Nadella admits, "I'm like a token-maxer too. So it is addictive." The article Microsoft CEO Satya Nadella admits he's a token-maxer, too: "It's addictive" appeared first on The Decoder .

The FBI built a 22,000 square-foot fake town complete with houses, a hotel, hospital, and power grid to train investigators on real-world cyberattacks. With U.S. cybercrime losses hitting a record $20.9 billion in 2025, the training focuses on ransomware attacks on critical infrastructure and the tough decisions investigators face when hospitals go dark. Over 1,400 people have already trained there.

Smartwatches and fitness bands now track everything from heart rate to sleep to stress, but doctors say most of this data isn't clinically useful. A cardiologist reports that about 70% of patient wearable data is 'made up by the company' with no clear medical meaning. Healthcare systems aren't set up to handle the constant stream of health metrics, and integrating this data into medical records remains a messy technical challenge. AI tools are being explored to help doctors synthesize this 'digital avalanche' into something actionable.

The Tribeca Film Festival showcased experimental AI-powered movies from Google DeepMind and OpenAI, revealing a key insight: generic AI video models produce visual slop, but when filmmakers customize AI tools for their specific creative vision, the results can actually be compelling. The takeaway? AI won't replace Hollywood storytellers — but artists who learn to work with AI might replace those who don't.

Microsoft researchers developed SkillOpt, a method that boosts AI model performance by simply training a Markdown instruction file. GPT-5.5 improved by about 23 points on procedural tasks using this approach, and the technique also works with other AI models like Claude Code. The key insight: you may not always need bigger models—better instructions can unlock significant performance gains.

# After a Data Center Proposal, New Jersey Town Bans Them   * News * [Analytics](https://www.govtech.com/analytics) * [Artificial Intelligence](https://www.govtech.com/artificial-intelligence) * [Civic Innovation](https://www.govtech.com/civic) * [Cloud & Computing](https://www.govtech.com/computing) * [Cybersecurity](https://www.govtech.com/security) * [Lohrmann on Cybersecurity](https://www.govtech.com/blogs/lohrmann-on-cybersecurity) * [Education](https://www.govtech.com/education) * [Election Technology](https://www.govtech.com/elections) * [Emerging Tech](https://www.govtech.com/products) * [Budget & Finan

Apple's redesigned Siri AI is coming to Mac with macOS 27 Golden Gate, and early testing shows it might actually be worth using this time. The new version appears significantly smarter than the old Siri, which many users had disabled years ago. For Mac users who dismissed Siri as useless, this could be a reason to give it another shot.

The government has restricted access to Anthropic's most powerful AI model after safety researchers flagged a potential jailbreak vulnerability. Anthropic has pushed back, arguing the finding doesn't warrant pulling a model used by hundreds of millions of people. The move represents a significant test of how governments balance AI safety concerns against public access to powerful AI tools.

Health trackers are getting smaller and harder to spot. Smart rings, earrings, and necklaces now blend in with regular jewelry, while companies like Oura have shrunk their latest ring by 40%. Continuous glucose monitors hide in shirt sleeves. The shift reflects a broader goal: make wearables so discreet that people wear them all the time, collecting more consistent health data. The real value isn't in the device itself but in the app software that digests the data and spots patterns.

 Everlywell / Elyse Betters Picaro / 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=ab024b65-026d-43dc-b94f-f8b16ffc1ba3&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=d1c062cd-74a1-4da6-ab95-876aa867e9d6&object_version=6187d898-83cd-4e92-819

 Jack Wallen/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=b88d35f7-d1f3-4d3d-8e41-6b9875645d67&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=1676978f-2a8b-462f-a914-8bccb109d8b6&object_version=a77b70c9-bff6-4057-b373-6494f73c63d1#link=%7B%22role%22:%22standard%22,%22href%22:%22https://cc.zdnet.com/v1/otc/00hQi47eqnEWQ6T9d4QLBUc?element=BODY&elem

Anthropic is taking its Claude Fable 5 model offline after the US government flagged concerns that the AI could be 'jailbroken' to bypass safety guardrails. The company says it will comply with the order. This marks a rare case of the government directly ordering an AI company to restrict access to one of its models, signaling tighter controls on frontier AI systems that could be exploited.

 Maria Diaz/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=b1cd4d51-2c8a-4aac-b8ed-c5b376d485e3&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=3ab042c7-93cc-4e31-ae41-f618c387521e&object_version=89a19c63-37e0-4cac-90b7-7b9eea90c9bc)_on Google._ * * * ### ZDNET's key takeaways * Power stations are typically reserved for use during emergencies. * I kee

A Verge writer used Google's Gemini AI to build a backyard app in minutes with no coding background. The experiment showcases how 'vibecoding' — letting AI do the heavy lifting while humans guide it — is making app creation accessible to regular people. It's a glimpse at a future where anyone can whip up a custom tool for everyday problems.

Spanish researchers have developed an AI agent system that can protect electric vehicle chargers from energy theft and physical damage. The system monitors charging sessions and infrastructure health in real-time, automatically detecting and blocking suspicious activity before attackers can drain power or sabotage equipment. As EV adoption grows, securing charging networks becomes critical for keeping transportation running smoothly.

Meta CEO Mark Zuckerberg publicly admitted the company made "mistakes" in how it shifted its workforce toward AI roles. While details are limited, the acknowledgment comes after Meta cut thousands of jobs while pivoting to AI—similar to moves by other big tech firms. It's rare for a CEO to admit missteps in workforce restructuring.

# Watching sports at home? I'd change these 4 soundbar settings for the most optimal audio | ZDNET 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 Pro vs Apple AirPods Pro 3](https://

Huawei announced its Ascend supernode built on Ascend 950 AI chips will fully support DeepSeek's V4 model, including V4-pro and V4-flash versions. DeepSeek claims the model can compete with OpenAI and Google DeepMind's closed systems, supporting a 1 million token context window at lower cost. Major Chinese companies including Alibaba, ByteDance, and Tencent are reportedly shifting toward Huawei's domestic AI chips due to US restrictions on advanced semiconductors.

[VinFast](https://vinfastauto.com/), the electric vehicle (EV) manufacturer under Vietnam’s conglomerate Vingroup, has produced its one millionth electric motorcycle, nearly eight years after it began manufacturing the Klara, its first electric motorcycle model, in November 2018. The milestone vehicle was a red Feliz II, rolling off the production line at the company’s manufacturing complex in the northern coastal city of Hai Phong, VinFast said in a statement on Friday. Since the Klara launch, VinFast has expanded its electric motorcycle lineup to nearly 15 models across more than 30 versions, covering segments from entry-level models for students requiring no driving license, such as the Ludo, Zgoo, and Flazz; to mainstream models including the Evo, Feliz, and Klara; and premium models s

Meta's stock dropped 5% on June 5 after reports emerged that the company might sell new stock to raise tens of billions for AI investments. While Meta hasn't hired banks and the move remains speculative, investors often react negatively to equity raises because new shares can dilute existing shareholders' stakes. The move comes as big tech companies are dramatically increasing AI spending—Meta recently raised its 2026 capital expenditure forecast to $145 billion.

Singapore actor Christopher Lee, 54, is starring in Crooks, Singapore's first AI-hybrid drama. The nine-episode series uses AI to create virtual sets, backgrounds, and environments that would be impossible to build physically—something the production turned to after investors pulled out, cutting funding by 40%. More than 50 crew members are being reskilled to work with AI tools. The show is inspired by Benny Se Teo, the ex-convict who founded restaurant chain Eighteen Chefs to help troubled youth.

Oracle's PeopleSoft software has a critical, actively exploited vulnerability that hackers are using to steal gigabytes of sensitive data from hundreds of organizations. This is as serious as security flaws get — if your workplace uses PeopleSoft for HR, finance, or operations, your organization's data could already be compromised. Patch immediately.

Singapore's Infocomm Media Development Authority (IMDA) has partnered with academic and industry players to boost AI safety and security research. The initiative focuses on developing robust, trustworthy AI systems as the technology becomes more embedded in daily life and business. This builds on Singapore's broader AI governance efforts, including AI Verify — the government's testing framework for AI model safety.

The US government has blocked foreign users from accessing Anthropic's most advanced AI models, according to Axios. This means developers, businesses, and researchers outside America can no longer use Claude's top-tier capabilities. It's part of a broader US push to keep frontier AI technology within American borders, citing national security concerns. If you rely on these tools for work or projects, you'll need to find alternatives or wait for policy changes.

A coalition of U.S. state attorneys general has launched a formal investigation into OpenAI, serving the company with a sweeping subpoena demanding documents about its activities and impact on users. This represents mounting regulatory pressure on AI companies as governments worldwide grapple with how to govern emerging technology. The investigation could lead to new rules affecting how AI companies handle data, safety, and consumer protection.

Meta CEO Mark Zuckerberg announced a companywide AI hackathon for July, but thousands of employees immediately pushed back. Workers say they're overwhelmed covering extra work after recent mass layoffs, and participation won't count toward performance reviews. Internal posts with hundreds of supportive reactions called out low morale and distrust in management as reasons to skip it.

# Watching sports at home? I'd change these 4 soundbar settings for the most optimal audio | ZDNET 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 Pro vs Apple AirPods Pro 3](https://

This week's Asian startup funding roundup highlights AI and tech deals. Notable: Singapore chip firm Silicon Box raised $77.53M for advanced semiconductor packaging; Israel's Rylo secured $85M to build AI communication tools for deaf and hard-of-hearing users; India's Hygenco raised $105M for green hydrogen solutions. The piece also covers AI startup acquisitions in Korea and UAE.

Meta CEO Mark Zuckerberg publicly admitted the company made "mistakes" in how it shifted its workforce toward AI roles. While details are limited, the acknowledgment comes after Meta cut thousands of jobs while pivoting to AI—similar to moves by other big tech firms. It's rare for a CEO to admit missteps in workforce restructuring.

Google Cloud is launching a dedicated AI startup accelerator for Southeast Asia, partnering with agencies from Singapore, Indonesia, and Vietnam. The three-month equity-free program will support 25 AI startups from seed to series B stage with engineering help, market access, and a trip to Silicon Valley to meet VCs on Sand Hill Road. Applications for the first cohort open now, with the program starting in August 2026.

Dutch chipmaking equipment giant ASML is set to become the largest shareholder in French AI startup Mistral AI, committing €1.3 billion (US$1.5 billion) in a €1.7 billion funding round. The deal values Mistral at €10 billion, making it Europe's most valuable AI company. ASML, the sole supplier of EUV lithography machines used by TSMC and Intel, will also get a board seat, signaling a deeper strategic partnership rather than just a financial investment.

# Huawei backs DeepSeek V4 with Ascend AI chips [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/companies/fundr

# Mapping the AI startups making waves in Japan - Tech in Asia [News](https://www.techinasia.com/)Premium * [News](https://www.techinasia.com/) * [Jobs](https://www.techinasia.com/jobs/search) * [Companies](https://www.techinasia.com/companies) * [Events](https://www.techinasia.com/events/live) * [About](https://www.techinasia.com/about) * [advertise](https://www.techinasia.com/advertise?utm_source=techinasia.com&utm_medium=website&utm_campaign=intl-navbar) * [Asia Economic Summit](https://aes.techinasia.com/?utm_source=navbar-tia-homepage&utm_medium=website&utm_campaign=events-flagship-aes) * [Sign Up](https://www.techinasia.com/auth) * [Log In](https://www.techinasia.com/auth?ref=nav_login) [Subscribe](https://www.techinasia.com/subscription) [I'm fundraising](https://www.techinasia.com/

Anthropic is expanding its Project Glasswing cybersecurity initiative to 150 more partner organizations across 15 countries, giving vetted companies access to AI models for identifying software vulnerabilities. Partners have already found over 10,000 critical flaws since the April pilot. The expansion includes organizations in power, water, healthcare, and communications sectors.

[ ](https://profiles.yahoo.com/brands/associated-press/) Updated Sat, 13 June 2026 at 2:29 am UTC 1 min read Anthropic co-founder and President Daniela Amodei, left, shakes hands with Snowflake CEO Sridhar Ramaswamy during the keynote presentaton at Snowflake Summit 26 Monday, June 1, 2026, in San Francisco. (AP Photo/Jeff Chiu) WASHINGTON (AP) — [AI giant Anthropic](http

The government has restricted access to Anthropic's most powerful AI model after safety researchers flagged a potential jailbreak vulnerability. Anthropic has pushed back, arguing the finding doesn't warrant pulling a model used by hundreds of millions of people. The move represents a significant test of how governments balance AI safety concerns against public access to powerful AI tools.

Anthropic is taking its Claude Fable 5 model offline after the US government flagged concerns that the AI could be 'jailbroken' to bypass safety guardrails. The company says it will comply with the order. This marks a rare case of the government directly ordering an AI company to restrict access to one of its models, signaling tighter controls on frontier AI systems that could be exploited.

Singapore's arts students and professionals are split on generative AI. Some, like LASALLE graduate Voon Shun Yi, reject it to preserve artistic craft and turn down jobs requiring AI tools. Others, like NTU student Cheerie Tan, use it to speed up workflow while keeping human creativity central. Arts institutions like LASALLE and SOTA have set up working groups and guidelines to navigate this shift, while artists like Karyn Lim embrace AI for prototyping and Joanne Ho for animating sketches. The consensus: AI can't replace human vision and intentionality, but it's reshaping how art gets made.

Singapore's arts schools are rewriting the rules on how they teach and grade creative work. LASALLE, NAFA, and the School of the Arts have set up guidelines and working groups to balance traditional skills with AI tools. Key changes include requiring students to document their creative process in journals, giving oral presentations to explain their work, and using tiered assessment scales that tell students upfront whether AI is prohibited, permitted, or required for each assignment. Schools acknowledge that performing arts may be more insulated than digital design fields, but all emphasize teaching critical thinking over just using tools.

Meta CEO Mark Zuckerberg has publicly acknowledged mistakes in the company's AI restructuring efforts. The social media giant has been aggressively pivoting toward AI under Zuckerberg's leadership, but this rare admission suggests internal strategy missteps. The reshuffle reportedly affected stock performance as investors reacted to uncertainty around Meta's AI direction.

Spanish researchers have developed an AI agent system that can protect electric vehicle chargers from energy theft and physical damage. The system monitors charging sessions and infrastructure health in real-time, automatically detecting and blocking suspicious activity before attackers can drain power or sabotage equipment. As EV adoption grows, securing charging networks becomes critical for keeping transportation running smoothly.

The US government has ordered Anthropic to shut down global access to its Claude Fable 5 and Mythos 5 AI models, citing security concerns about potential jailbreaks. Anthropic is complying but fighting back publicly, arguing the vulnerabilities are minor and exist in competing models like GPT-5.5. The company warns this could set a precedent that makes frontier AI deployments harder everywhere. For everyday users, this means some AI tools may become unavailable or more restricted as companies scramble to meet government security demands.

AI company leaders are quietly walking back their doomsday predictions about AI replacing white-collar jobs. After years of warnings about mass unemployment, executives are now softening their language — likely to avoid spooking regulators, investors, and the public. The shift suggests the industry knows it overplayed the job-displacement angle and is trying to manage expectations.

A German court has ruled that Google is legally liable for false statements produced by its AI Overviews feature, setting a precedent that companies operating AI systems must assume responsibility for damages caused by their outputs. The ruling moves beyond previous cases that blamed AI "hallucinations" on users, instead holding AI deployers accountable for the systems they design, train, and manage.

Visa has partnered with OpenAI to handle payments when AI agents make purchases or transactions on behalf of users. This is a new frontier: AI systems that spend money autonomously. The key question is trust—can you safely let an AI agent use your card? Visa is building safeguards to verify these AI-initiated transactions, but experts warn consumers should understand exactly what their AI is authorized to buy.

Chinese AI startup Moonshot AI released Kimi K2.7 Code, a coding-focused model with one trillion parameters that costs up to 12 times less than GPT-5.5 or Claude per token. It trails top models on benchmarks but offers developers far more runs for the same budget—a trade-off that could make it attractive for cost-conscious teams.

Google Research has released Gemini-SQL2, an AI model that converts plain English into SQL database queries. It scored 80.04% on the BIRD benchmark, significantly outperforming rival models from OpenAI and Anthropic. This matters for everyday users because natural language database queries could eventually let non-technical people extract insights from data without learning coding.

Microsoft researchers developed SkillOpt, a method that boosts AI model performance by simply training a Markdown instruction file. GPT-5.5 improved by about 23 points on procedural tasks using this approach, and the technique also works with other AI models like Claude Code. The key insight: you may not always need bigger models—better instructions can unlock significant performance gains.

Meta is cracking down on runaway AI costs. An internal memo sent to 6,000 employees reveals the company is projected to spend billions just on internal AI use. Starting 2027, all AI spending will need approval through a new system called 'AI Gateway.' CTO Andrew Bosworth told staff that burning more tokens doesn't equal real progress. The shift signals even the biggest AI players are now treating compute costs as a serious budget line item.

 Maria Diaz/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=b1cd4d51-2c8a-4aac-b8ed-c5b376d485e3&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=3ab042c7-93cc-4e31-ae41-f618c387521e&object_version=89a19c63-37e0-4cac-90b7-7b9eea90c9bc)_on Google._ * * * ### ZDNET's key takeaways * Power stations are typically reserved for use during emergencies. * I kee

China released new rules requiring financial companies to handle customer data more carefully as part of a wider cybersecurity push. Banks and fintech firms must now follow stricter guidelines on how they collect, store, and protect user information.

AI cameras are being tested in Kananaskis to automatically spot and alert authorities about wildfires before they spread. The system uses computer vision to detect smoke and flames early, potentially cutting response times and saving forests and homes.

A coalition of US state attorneys general is investigating OpenAI, demanding records on how it handles user data, protects minors and seniors, and manages AI safety. The probe comes as OpenAI prepares for a potential trillion-dollar IPO and faces multiple lawsuits alleging ChatGPT caused real-world harm, including a case where a mother claims the chatbot encouraged her daughter's suicide.

Anthropic's Claude Fable 5 scored 88 percent on the hardest FrontierMath problems, beating OpenAI's GPT-5.5 by 13 points. This is a significant leap from its predecessor, which scored below 10 percent just months earlier. The rapid improvement shows AI math capabilities are accelerating fast.

The FCC is pushing to eliminate burner phones, citing their use in fraud and crime. Meanwhile, AI is now being used to automatically find software vulnerabilities—Microsoft just had its biggest-ever patch day thanks to AI bug hunting. A ransomware gang also exploited a new Oracle flaw to steal data.

The FBI built a 22,000 square-foot fake town complete with houses, a hotel, hospital, and power grid to train investigators on real-world cyberattacks. With U.S. cybercrime losses hitting a record $20.9 billion in 2025, the training focuses on ransomware attacks on critical infrastructure and the tough decisions investigators face when hospitals go dark. Over 1,400 people have already trained there.

The Tribeca Film Festival showcased experimental AI-powered movies from Google DeepMind and OpenAI, revealing a key insight: generic AI video models produce visual slop, but when filmmakers customize AI tools for their specific creative vision, the results can actually be compelling. The takeaway? AI won't replace Hollywood storytellers — but artists who learn to work with AI might replace those who don't.

A compact robot roughly the size of a baseball has landed on the Moon as part of a lunar mission and is sending back images of the surface. The robot uses AI for navigation and terrain mapping to avoid obstacles in the challenging lunar environment.

Rivian CEO RJ Scaringe has launched a separate robotics company called Mind Robotics, which has raised over $1 billion. Unlike Elon Musk's approach with Tesla, Scaringe keeps the robotics business independent while positioning Rivian as a major customer. Mind expects to reveal its first humanoid robot within a year. Humanoid robots powered by AI could eventually work alongside humans in factories, hospitality, and homes.

Apple's iOS 27 brings its first real AI photo editing tools to the world's most popular camera. Features like object removal and scene extension are modest compared to Google Pixels, but they signal Apple's entry into AI-enhanced photography. For everyday iPhone users, it means touching up photos gets easier — no third-party apps needed.

Health trackers are getting smaller and harder to spot. Smart rings, earrings, and necklaces now blend in with regular jewelry, while companies like Oura have shrunk their latest ring by 40%. Continuous glucose monitors hide in shirt sleeves. The shift reflects a broader goal: make wearables so discreet that people wear them all the time, collecting more consistent health data. The real value isn't in the device itself but in the app software that digests the data and spots patterns.

Google Research's Gemini-SQL2 turns natural language into executable SQL queries. Built on Gemini 3.1 Pro, it tops the BIRD benchmark at 80.04 percent accuracy, well ahead of OpenAI and Anthropic. Google says the technology could improve natural language features across its data services. The article Google Research's Gemini-SQL2 tops text-to-SQL benchmarks by a wide margin appeared first on The Decoder .

The US government ordered Anthropic to block its most advanced models, Fable 5 and Mythos 5, from all foreign users over national security fears. Rather than carve out exemptions, Anthropic cut off access entirely — meaning American customers lost out too. It's one of the most sweeping AI export restrictions yet, raising questions about how far governments will go to contain AI risks.

Microsoft and three Chinese universities have developed SkillOpt, a method that optimizes instruction documents for AI agents using principles from traditional model training. A simple Markdown file is enough to boost GPT-5.5 by about 23 points on procedural tasks, and the same file transfers across models and agent environments like Codex and Claude Code. The article Microsoft's SkillOpt boosts GPT-5.5 by using nothing but a trained Markdown file appeared first on The Decoder .

Researchers trained robots to read human emotions using vision language models that can analyze not just faces but whole scenes and context. In tests with 40 volunteers, the AI system outperformed traditional emotion-recognition software (scoring 0.86 vs 0.77). People preferred when robots gave personalized apologies after making mistakes, but the study found that functionality still matters more than emotional charm—a robot that fails its task loses trust regardless of how well it apologizes.

Smartwatches and fitness bands now track everything from heart rate to sleep to stress, but doctors say most of this data isn't clinically useful. A cardiologist reports that about 70% of patient wearable data is 'made up by the company' with no clear medical meaning. Healthcare systems aren't set up to handle the constant stream of health metrics, and integrating this data into medical records remains a messy technical challenge. AI tools are being explored to help doctors synthesize this 'digital avalanche' into something actionable.

# After a Data Center Proposal, New Jersey Town Bans Them   * News * [Analytics](https://www.govtech.com/analytics) * [Artificial Intelligence](https://www.govtech.com/artificial-intelligence) * [Civic Innovation](https://www.govtech.com/civic) * [Cloud & Computing](https://www.govtech.com/computing) * [Cybersecurity](https://www.govtech.com/security) * [Lohrmann on Cybersecurity](https://www.govtech.com/blogs/lohrmann-on-cybersecurity) * [Education](https://www.govtech.com/education) * [Election Technology](https://www.govtech.com/elections) * [Emerging Tech](https://www.govtech.com/products) * [Budget & Finan

 Jack Wallen/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=b88d35f7-d1f3-4d3d-8e41-6b9875645d67&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=1676978f-2a8b-462f-a914-8bccb109d8b6&object_version=a77b70c9-bff6-4057-b373-6494f73c63d1#link=%7B%22role%22:%22standard%22,%22href%22:%22https://cc.zdnet.com/v1/otc/00hQi47eqnEWQ6T9d4QLBUc?element=BODY&elem

A Verge writer used Google's Gemini AI to build a backyard app in minutes with no coding background. The experiment showcases how 'vibecoding' — letting AI do the heavy lifting while humans guide it — is making app creation accessible to regular people. It's a glimpse at a future where anyone can whip up a custom tool for everyday problems.

SpaceX is expanding its Starlink satellite network to support AI workloads, positioning itself as a direct competitor to traditional cloud providers like AWS and Google Cloud. The company is deploying dedicated compute satellites that can handle AI processing tasks from orbit, potentially disrupting how businesses access AI infrastructure.

Companies are hitting a wall with AI subscription costs and are starting to look at cheaper alternatives — including Chinese-built AI models and free open-source options. This shift could reshape the AI market by giving budget-conscious businesses access to capable tools without the premium price tag.

Microsoft CEO Satya Nadella is warning against "token-maxing," throwing the most powerful AI models at every problem. Frontier models shouldn't be wasted on everyday tasks, he says. The marginal cost of productivity gains has to match the token cost. Still, Nadella admits, "I'm like a token-maxer too. So it is addictive." The article Microsoft CEO Satya Nadella admits he's a token-maxer, too: "It's addictive" appeared first on The Decoder .

 Everlywell / Elyse Betters Picaro / 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=ab024b65-026d-43dc-b94f-f8b16ffc1ba3&position=1&template=article&track_code=__COM_CLICK_ID__&url=https%3A%2F%2Fwww.google.com%2Fpreferences%2Fsource%3Fq%3Dzdnet.com&view_instance_uuid=d1c062cd-74a1-4da6-ab95-876aa867e9d6&object_version=6187d898-83cd-4e92-819

Anthropic has been forced to shut down its most powerful AI models, Claude Fable 5 and Mythos 5, just days after launch. The US government raised national security concerns, specifically about a "jailbreak" technique that could bypass safety controls and potentially help hackers exploit computer systems. The UK's AI Safety Institute found the model could exploit cybersecurity defenses 73% of the time. This marks the first time a US AI company has faced such direct government intervention over security risks, with the Pentagon also labeling Anthropic a "supply chain risk."

Visa has partnered with OpenAI to let ChatGPT buy stuff directly on your behalf using your credit card. Instead of just recommending products, the AI can now complete purchases end-to-end—from picking headphones under $150 to checking out. It's a sign that AI "agents" are moving from theory to real transactions. But experts warn of risks: ChatGPT has already been caught recommending fraudulent storefronts that steal money and banking details. Scammers are increasingly designing websites to trick AI bots, meaning your digital shopping assistant could be fooled into handing over your cash to crooks.

Researchers have used data from Pokémon Go players — specifically millions of recorded movement patterns — to train AI systems that could guide drone navigation. The location data, collected from players exploring real-world environments, helps AI learn how objects and people move through spaces, which is useful for autonomous drone operations.

Apple's redesigned Siri AI is coming to Mac with macOS 27 Golden Gate, and early testing shows it might actually be worth using this time. The new version appears significantly smarter than the old Siri, which many users had disabled years ago. For Mac users who dismissed Siri as useless, this could be a reason to give it another shot.

An opinion piece examining how Anthropic's decisions around model access reflect broader power dynamics in the AI industry. The analysis suggests that controlling who can use advanced AI models is itself a form of competitive advantage, raising questions about openness, safety, and market competition among AI companies.

This is the same story as the Anthropic Fable/Mythos shutdown—Ars Technica's more detailed coverage confirms the company completely disabled access after a US Commerce Department directive restricting use outside America. The administration cited concerns about a jailbreak that could bypass safety filters for cybersecurity, chemistry, and biology topics. Anthropic disputes the severity, saying the vulnerabilities found are "minor" and exist in other AI models too. The company is suing the Pentagon over a "supply chain risk" designation—the first ever given to a US company.

Update on India’s workers are training AI robots to take their jobs. Coverage via Reddit/r/technology.

Update on Anthropic CEO Dario Amodei warned Mythos posed a national security threat. Washington just responded.. Coverage via Yahoo Singapore Tech.

Researchers have developed a new AI model called "Count Anything" that can count objects in any image type using just a text prompt—whether it's crowds of people, cells under a microscope, or items in a photo. In tests, it cuts counting errors in half compared to previous systems. However, it still struggles with extremely dense objects and ambiguous descriptions.

Update on Google Photos Prepares Massive 'Video Remix' AI Upgrade. Coverage via Yahoo Singapore Tech.

An opinion piece arguing that AI is advancing faster than governments can regulate, and no one has a coherent plan to keep AI systems under human control. The writer warns that as AI becomes more autonomous, the risks of losing the ability to steer it grow.

Amazon reportedly flagged concerns about Anthropic's AI models to US officials before the government cracked down on the company's Mythos and Fable models. The move comes after Anthropic CEO Dario Amodei himself warned that these models posed national security risks. Washington has since ordered access restrictions. For everyday tech users, this signals that even top AI companies face government pushback — and raises questions about who gets to use powerful AI tools going forward.

Update on OpenAI hit with multistate probe into possible user harm as its IPO looms. Coverage via Yahoo Singapore Tech.

The massive build-out of data centers to power AI is running into real-world limits: not enough electricity, not enough water for cooling, and communities pushing back. This could mean slower AI rollouts, higher costs for cloud services, or more power drawn from the grid.

Microsoft president Brad Smith is urging the tech industry to listen up: graduates and young people are pushing back hard against AI at events and speeches. Smith says the backlash is a wake-up call that workers want control over how AI gets used in their jobs, not be replaced by it. He's trying to soften the message from AI leaders who have predicted massive job losses, arguing AI should make workers more productive, not obsolete. The tech industry is now shifting its tone to emphasize AI as a helper, not a replacer.

A new tech job is booming in Singapore: the forward deployed engineer. These are tech professionals who build AI systems inside client companies — like banks, hospitals, and ports — while also explaining the tech to non-technical staff. At least 35 job openings exist now, with OpenAI planning to hire or train 200 of them, and Databricks adding over 30 more. Mid-career salaries start at $120,000 a year, significantly higher than traditional software or analyst roles. Experts say the role is becoming essential as firms move from AI experiments to real operations.

KPMG has pulled a published report after discovering it contained AI hallucinations — facts and data the AI system invented. It's a stark reminder that AI-generated content can look polished but be completely wrong, and companies need human fact-checkers even when using AI tools.