Singapore AI news archive.
Every SG + AI news item we classified, grouped by ISO week. Each card carries the analyst one-liner, category tag, and sentiment dot that appeared on the live reel. Kept permanently so we can look back on how the ecosystem evolved.
About this archive: each item is ingested and classified automatically every day with a category (policy, enterprise, funding, talent, infrastructure or products), a sentiment read, and a one-line analyst note. The archive is kept permanently. The same feed powers the news sections on the altronis.sg homepage and the @AIinSG Telegram channel.
Week 43 · 2026
60 items
ProductsWIR · The PHP Meltdown Shows How AI Makes Bugs More Exploitable
AI turned dormant code flaws into immediate threats, forcing defenders to patch at machine speed or face mass exploitation.
ProductsDEC · Fancy harnesses make coding agents more expensive but not better, study finds
Harness switching barely moves pass rates but can triple costs, exposing that vendor benchmarks often measure noise rather than capability.
ProductsTC-AI · Hottest toy for Christmas? Nvidia CEO plays with adorable Microduck robots, 20,000 sell
Nvidia's Hugging Face acquisition is already shipping hardware, proving the M&A strategy extends beyond software to physical AI.
ProductsVERGE · OpenAI now trains its frontier AI models with no internet access
Offline training is now the default for frontier models, forcing a rethink of how we test agent containment.
ProductsVERGE · Jack Dorsey’s Block made an AI model that talks to you like an awkward new friend
Block's speech-native model challenges the text-first pipeline, proving local compute can handle nuanced voice without cloud latency.
ProductsDEC · Chinese AI model Kimi K3.1 Pro is closing in on Anthropic's Claude Opus 5.5
The 3-point gap to proprietary leaders forces a re-evaluation of open-weights viability for high-stakes agentic workflows.
ProductsQCR · RhymeQ Joins NVIDIA Inception Program and Announces Technical Collaboration on NOOA Agentic AI Framework
Agentic AI is now being wired directly into quantum hardware orchestration, a signal that hybrid compute stacks are becoming a first-class engineering target.
ProductsDEC · Anthropic says its unreleased model is both more capable and better aligned than Mythos 5.1
Anthropic's internal model halting 67% of rogue attacks suggests alignment is becoming a measurable, quantifiable metric rather than just a safety checkbox.
ProductsHF · LitPivot: Developing research ideas through dynamic contextualization within the literature landscape
Ai2's LitPivot pairs LLM critique with literature retrieval, showing that grounded ideation tools outperform generic chat for research rigor.
ProductsVERGE · Retro gaming is being overrun with AI
AI coding tools are compressing years of reverse-engineering into weeks, forcing a reckoning on whether speed undermines the preservationist ethos.
ProductsDEC · Canva launches Craft, its first in-house image model
Canva's in-house model shifts the design tooling market from API dependency to vertical integration, prioritizing speed and cost control.
ProductsDEC · Anthropic's Claude now writes code and proves it safe automatically
Autonomous formal verification collapses the cost of proving code correctness, shifting the bottleneck from human labor to compute.
ProductsTQI · Google Launches $150,000 AI Competition to Reduce Quantum Computing Demands
Google is betting on community-driven optimization to shrink quantum circuit overhead, a key bottleneck for practical utility.
ProductsQCR · Google Quantum AI Launches $150K Kaggle Competition to Compress Fault-Tolerant Quantum Circuits Using AI Agents
Google is betting that agentic AI can crack the compilation bottleneck for fault-tolerant quantum, turning a niche optimization problem into a public benchmark.
ProductsQCR · Leiden University and Riverlane Research Reveals Distance-Losing Flaw in Correlated Decoding Below Threshold
Fixing a silent distance-loss bug in correlated decoders removes a major scaling bottleneck for surface-code hardware.
ProductsSCMP · Asia-Pacific AI agent market to surge 14-fold to US$11.2 billion by 2030, Omdia says
Omdia projects a 14-fold surge in APAC AI agent revenue to $11.2B by 2030, driven by Chinese tech firms deploying automated tools.
ProductsQCR · IQM and NVIDIA Automate 20-Qubit Chip Calibration via NVIDIA Ising Vision-Language Models and QCalEval Benchmark
Automating qubit calibration with VLMs removes the manual engineering bottleneck, making the path to scaling superconducting hardware significantly more viable.
ProductsTNG · Google Cloud to offer OpenAI models through Model Garden
Unified procurement removes the biggest friction for enterprises running multi-model agent stacks, though preview status leaves compliance gaps.
ProductsTC-AI · Left alone with nothing to do, AI agents started playing
Unstructured play emerged as a viable training signal, suggesting agents can self-generate the curriculum needed for complex robotic control.
ProductsHF · Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox
Microsoft's shift from static grading to executable database-state rewards suggests a new standard for verifying agent reliability in production.
ProductsVERGE · Canva’s first AI image model plans designs before generating them
Canva's two-stage planning architecture targets the precision gap in design tools, positioning it as a direct challenger to OpenAI.
ProductsMIT · Designing an AI serving harness built to last
The competitive moat is shifting from model selection to the serving harness, forcing enterprises to optimize for latency and cost rather than raw capability.
ProductsVERGE · Meta turns Muse sharing back on
Meta patched a DNS enumeration flaw in Muse that exposed private business data, forcing a reset of public links.
ProductsTC-AI · 1Password launches a way for AI agents to log in on your behalf without seeing your password
Credential isolation for agents is now a product category, forcing banks to rethink how they verify automated user intent.
ProductsTC-AI · Canva releases its first image generation model
Canva's native model targets API cost reduction and design fidelity, signaling a shift from third-party reliance to proprietary control.
ProductsOAI · Build more responsive apps with the Decisions API
OpenAI is narrowing the API surface to cut latency and cost, pushing developers toward lightweight logic over heavy reasoning.
ProductsTC-AI · OpenAI makes its Jev competitor widely available as decision model competition heats up
OpenAI's Decisions API targets the sub-200ms inference window, forcing a rethink of LLM costs for high-throughput classification tasks.
ProductsTC-AI · Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’
Microsoft's agent orchestration adds little value over raw model power, suggesting the 'medical superintelligence' gap is closing via base model upgrades.
ProductsVERGE · In Meta’s version of the future, I’ll have my Muse call your Muse
Meta is betting its ad model on agent-to-agent commerce, but the lack of concrete retention data leaves the revenue thesis unproven.
ProductsVERGE · Microsoft says its AI tool is better than doctors at diagnosing conditions
Microsoft's multi-agent diagnostic tool outperforms physicians by 4x, signaling a shift from single-model reliance to orchestrated reasoning.
ProductsVERGE · Someone hacked the Switch 2 by using an old game and AI
AI-assisted reverse engineering is lowering the barrier for console exploits, potentially accelerating the pace of jailbreaks.
ProductsTC-AI · Chinese hackers spied on thousands of companies after breaching IT giant ConnectWise
AI agents turned a single credential leak into a 3,000 sign-ins/minute assault, proving that automated tooling now dictates the speed of breach detection.
ProductsHF · OlmoEarth v1.3: State-of-the-art embeddings for the whole planet
Precomputed global embeddings lower the barrier for planetary-scale analysis, shifting the workload from heavy GPU training to lightweight local classification.
ProductsDEC · Recursive shows first result of its AI-driven research process with Flint
Recursive's Flint proves AI can design its own architecture, but the lack of independent verification keeps the self-improvement claim speculative.
ProductsGDM · Beyond tree cover: mapping the world’s forest types
High-res forest mapping is a benchmark for Earth observation models, validating the utility of AI for granular environmental data.
ProductsCNBC TV · 'You contain the baby and you monitor it': Nvidia's Huang on AI safety
Nvidia is positioning its agent containment tools as the necessary safety layer for autonomous AI, shifting the focus from model capability to operational control.
ProductsTC-AI · Claude Duet app is Anthropic’s bet on agents
Anthropic targets the SMB market with Duet, betting that autonomous task execution beats passive chat for professional workflows.
ProductsSCMP · Unitree developing AI system to let humanoid robots improve themselves
Unitree is automating the robotics dev loop, a move that could compress the timeline for physical AI deployment.
ProductsVERGE · 1Password lets AI agents use your passwords without seeing them
Credential isolation removes the biggest trust barrier for agent adoption, letting users delegate tasks without handing over permanent access.
ProductsVERGE · AI agents are ready to shop. Some retailers are wary.
Retailers are actively walled-gardening against agentic commerce, forcing a rethink of how AI agents interface with existing e-commerce infrastructure.
ProductsVERGE · Claude did not invent scorpion antivenom
The experiment exposes a hard ceiling for LLMs in de novo drug design, validating the need for specialized bio-tools.
ProductsQCR · qBraid Unveils qBraid OS and Launches Hosted GH200 GPU Simulators at NVIDIA GTC Berlin
Standardizing quantum control via GPU co-processing lowers the operational barrier for hybrid quantum-classical stacks.
ProductsDEC · An AI agent in "Skynet mode" hijacks unprotected language models to launch self-sustaining cyberattacks
LLMjacking turns exposed endpoints into attack vectors, forcing a hard look at API hygiene before the next breach hits.
ProductsVERGE · Eiliko maker says it’s hiring an ‘independent security firm’ after Verge report on its insecure robots
Eiliko’s rushed patch fails to stop live data leaks, proving that consumer AI hardware security is a critical, unsolved gap.
ProductsDEC · OpenAI agents analyze mouse brain data in 32 minutes, a job that would take humans months
Distributed agents outperforming single models in blind tests suggests the agent swarm is the new baseline for complex scientific reasoning.
ProductsWIR · Hugging Face’s New AI Biology Model Lets You Chat With DNA
Running locally on a laptop lowers the barrier for genomic analysis, shifting power from cloud-heavy labs to individual researchers.
ProductsDEC · Anthropic takes on Grok Bot, Meta's Muse, and OpenAI's Dots with its always-on agent "Claude Duet"
The shift to always-on agents forces a rethink of trust boundaries, as approval gates become the primary control mechanism.
ProductsHF · Ensemble forecasting with IBM Granite time-series models
IBM removes the model-selection tax for time-series forecasting by shipping a default ensemble recipe on Hugging Face.
ProductsST · Tesla aims to sell Cybercab robotaxis by end of 2026 despite federal probe
Tesla pushes a control-free vehicle into the market before regulators finish rewriting the rules, forcing a live test of self-certification limits.
ProductsTNG · Cloudera’s Carolyn Duby on governing AI-agent access and security [Q&A]
Security teams must shift from static identity checks to continuous blast-radius monitoring as agents gain write access to production systems.
Productse27 · AI agents broke into 100 firms for pocket change. SEA’s online shops fit the target profile
Autonomous agents breached 100 firms for pennies, exposing that SEA e-commerce security is effectively non-existent.
ProductsSCMP · Chinese firm unveils ‘guardian’ robot that can disarm attackers, lift cars
Unitree pivots from athletic demos to tactical utility, signaling a push for real-world deployment despite vague release details.
ProductsTC-AI · Meta says Muse has over 10M users, with shopping a top activity
Agentic commerce is hitting real revenue traction, with $200 average ticket sizes validating the model for high-value transactions.
ProductsTC-AI · Recursive launches Flint, a laptop coding model it says was largely built by AI
Recursive's Flint proves AI can build AI, but the real signal is that 40B MoE models now run locally, shifting dev workflows away from cloud.
ProductsVERGE · Anthropic takes on Muse with Claude Duet, its new AI agent
Anthropic targets the SMB market with a proactive agent, betting that daily briefs and task automation will drive retention.
ProductsDEC · OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon found critical PHP flaw affecting millions of websites
AI models now scale exploits across PHP versions, collapsing the effort gap between finding and weaponizing flaws.
ProductsVERGE · Walmart is inviting AI agents to shop for you
Walmart's MCP-based gateway signals that agentic commerce is shifting from browser scraping to structured API integrations.
ProductsTC-AI · While Amazon blocks AI agents, Walmart rolls out the welcome mat
Walmart’s MCP gateway signals that agentic commerce is moving from hype to hard infrastructure, forcing rivals to rethink access.
ProductsARS · OpenAI's latest AI models are coming to Google Cloud and Antigravity
OpenAI models landing on Google Cloud breaks the exclusivity barrier, forcing enterprises to treat model choice as a pure utility decision.
ProductsHF · NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required
NVIDIA's Kumo Relational eliminates feature engineering for structured data, forcing a rethink of legacy XGBoost pipelines.