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 34 · 2026
60 items
ProductsDEC · Deepseek releases experimental Flash vision model that rivals Opus 4.8 on agent benchmarks
DeepSeek expands V4-Flash with multimodal agent capabilities, challenging Opus 4.8 on internal benchmarks and lowering integration friction via API compatibility.
ProductsQCR · Quantum X Labs Outperforms PyMatching Benchmarks on Google Quantum Hardware Surface-Code Dataset Using NVIDIA CUDA-Q
AI-driven QEC decoding on CUDA-Q beats PyMatching on Google hardware, showing practical utility for quantum error correction.
ProductsSCMP · OpenAI-backed legal tech firm pivots to Chinese Kimi K3 open-weight model
Harvey Tenet post-trains on Kimi K3, showing open-weight models are viable for specialized enterprise workloads.
ProductsSO · Dispatches from O'Reilly: The right amount of spec for agentic development
Cheap code generation exposes specification gaps, forcing teams to rebuild formal checks to manage agentic output quality.
ProductsTQI · Quantum X Labs Tests AI Quantum Error Decoder on Google Hardware Dataset
AI-driven error decoding on Google hardware signals a practical path toward fault-tolerant quantum computing.
ProductsVERGE · Major YouTube creators are facing backlash for accepting AI money
Higgsfield's aggressive creator outreach backfires, exposing the friction between AI tooling and audience trust in creative sectors.
ProductsGDM · From Atari to EVE Online: Building on 15 Years of AI Research in Games
DeepMind shifts from benchmarking to co-creating with studios, signaling a push for embodied AI in complex, open-ended environments.
ProductsARS · As demand for Meta AI glasses explodes, it’s harder to avoid creepy recordings
Privacy backlash forces Meta to confront hardware detection gaps, complicating adoption for enterprise and public use cases.
ProductsCNA · Singapore government beta testing new chatbot to make official information more accessible
GovTech deploys a multilingual chatbot for public info, shifting routine queries away from human officers to improve service efficiency.
ProductsY! · ACE Robotics CEO says robot brains will have 'ChatGPT moment' by end of 2027
Embodied AI benchmarks shift focus to real-world utility, validating world models as the next inflection point for robotics.
ProductsAIN · How AI coding tools are contributing to the popularity of JavaScript
AI coding agents reinforce JavaScript dominance via training data bias, narrowing the effective dev stack rather than making it stack-agnostic.
ProductsST · New AI-powered system to help detect motorists flouting traffic rules like driving in bus lanes
LTA’s AI video analytics moves from trial to deployment, shifting enforcement from manual summonses to automated detection by 2026.
ProductsDEC · GPT-5.6 Sol drives OpenAI's revenue surge as it regains ground on Anthropic
OpenAI's new model drives a 35% revenue jump, signaling that frontier capability still dictates enterprise spending growth.
ProductsDEC · OpenAI's GPT-Image-2 can now generate images without a background
OpenAI adds transparent background generation to GPT-Image-2 API, simplifying asset creation for e-commerce and design workflows.
ProductsSO · Get rid of your CAPTCHA, the future of the web is bots
WordPress VIP argues AI agents are shifting web architecture, making structured content essential for bot interaction.
ProductsHF · Measuring benchmark optimization in speech recognition
Benchmark overfitting in ASR is real, forcing a shift from leaderboard scores to held-out evaluation sets for genuine capability assessment.
ProductsQCR · Crypto4A Achieves World-First FIPS 140-3 Level 3 Validation for Quantum-Safe HSM Module
First quantum-safe HSM hits FIPS 140-3 Level 3, validating PQC readiness for critical infrastructure ahead of migration deadlines.
ProductsTC-AI · ChatGPT can now send texts for you with new Apple Messages plug-in
OpenAI extends ChatGPT into iOS messaging, shifting the product from analysis to autonomous action with new privacy trade-offs.
ProductsVERGE · Google Discover is getting an AI chatbot-tuned feed
Google shifts Discover personalization to natural language, signaling a broader industry pivot toward chat-driven content curation interfaces.
ProductsDEC · Adobe Firefly adds AI audio tools and Google's Gemini Omni Flash
Adobe integrates Gemini Omni Flash and audio generation, expanding Firefly's multimodal capabilities for commercial content creation.
ProductsTC-AI · Google gives publishers a new way to fight AI-driven traffic losses
Google expands publisher control to mitigate AI search traffic erosion, signaling a shift toward user-curated source attribution.
ProductsTC-AI · Runlayer, Rippling drop lawsuits — but the brouhaha is still a cautionary tale for founders
Rippling’s immediate MCP gateway launch after dropping the suit signals that incumbents will clone AI tooling rather than partner.
ProductsTC-AI · Linkdaze’s smart calendar is built to run a household, not just track a schedule
Hardware vendor bets on upfront pricing and AI meal planning to differentiate from subscription-heavy smart displays.
ProductsDEC · LLMs could write like humans but post-training guardrails make their text detectable
Guardrail-induced mode collapse narrows LLM style, making post-trained outputs easier to detect than raw base models.
ProductsTC-AI · Grok keeps sending gibberish responses to users
Grok Lite generation glitches highlight stability risks as xAI scales output, potentially eroding trust in its efficiency claims.
ProductsMS · You can ask humanoid robot at Little India MRT station for directions from Aug. 31-Sep. 11
Humanoid deployment in public transit signals a shift from lab demos to real-world operational testing in SG.
ProductsTC-AI · A third of web pages published since ChatGPT’s launch show signs of AI authorship, study finds
AI-generated content now dominates new web pages, forcing a structural shift in search and content verification strategies.
ProductsTC-AI · Ramp launches its own AI model router, called Router
Ramp enters the model routing layer, signaling that inference management is becoming a standard feature for enterprise SaaS platforms.
ProductsTC-AI · Meta brings Pocket, an app that lets you vibe-code and share games, to US users
Meta expands its AI creation tooling portfolio with a US rollout, signaling a push to mainstream AI-generated interactive content.
ProductsAIN · Amazon’s Prime Air autonomous drones to reach 500 US cities
Autonomous flight software scales Prime Air to 500 US cities, proving detect-and-avoid systems can replace remote pilots at volume.
ProductsVERGE · Welcome to the AI crisis in math
Frontier models solving abstract math signals capability leaps, forcing a re-evaluation of academic research value and grant structures.
ProductsARS · Grok exfiltrates user data when malicious instructions are encrypted
Grok's safety guardrails fail against encrypted injection, exposing user data and highlighting persistent LLM vulnerability vectors.
ProductsDEC · GEN-1.5: Generalist AI teaches robots new tasks from a single demo
Generalist AI's GEN-1.5 enables zero-shot robot task learning from single demos, signaling a shift toward in-context physical prompting.
ProductsDEC · KI-Pioneer Sutton calls synthetic data a "big mistake" in the face of an infinitely complex world
Sutton rejects synthetic data as a scaling path, signaling a pivot toward continuous agent learning over static model training.
ProductsTC-AI · Meta AI’s new Mac app wants you to talk to your apps
Meta extends Muse Spark to Mac for system-wide dictation and business insights, pushing agents into daily workflows.
ProductsVERGE · Slack is launching collaborative vibe coding channels
Slack embeds coding agents into chat, shifting AI dev tools from isolated apps to collaborative workflows.
ProductsDEC · Anthropic's most capable model, codenamed "Model 2," is for internal use only
Anthropic's internal Model 2 shows incremental gains over public Claude, suggesting a cautious release cadence rather than a leapfrog capability jump.
ProductsTC-AI · Binance now lets AI agents trade, but keeping them in check is largely up to users
Binance launches Agent OS for autonomous trading, shifting safety liability to users via sub-account sandboxing.
ProductsTNG · Eight Korean AX companies will run the Korea Pavilion, ‘X- DIGITAL@THAILAND,’ at Techsauce Global Summit 2026
Korean AI firms targeting Thailand signals regional expansion pressure, shifting competition from domestic markets to Southeast Asian adoption.
ProductsTNG · How to spot phishing when AI removes the old warning signs
AI-polished phishing hits 56% of samples, forcing a shift from syntax checks to behavioral analysis in security stacks.
ProductsAIN · A third of ChatGPT ads appear in irrelevant conversations
OpenAI's ad relevance engine is underperforming, suggesting the targeting model needs significant tuning before monetization scales.
ProductsDEC · OpenAI builds safety system that catches misuse without storing customer data
OpenAI's zero-retention safety signal lets enterprises use frontier models without data leakage, tightening the compliance moat.
ProductsAIN · HoneyBook bets on agentic AI to streamline small business operations with its new Claude connector
HoneyBook extends agentic AI to SMEs via MCP, closing the enterprise adoption gap for smaller firms.
ProductsSCMP · Hesai rides on lidar growth as smart cars gain ground despite China’s struggling market
Lidar adoption in China accelerates, validating sensor-heavy autonomy stacks over vision-only approaches.
ProductsY! · China puts robocops on traffic duty, minus the arrest powers
China's SUPCON deploys autonomous traffic robots, proving AI-driven physical automation can handle real-world enforcement at scale.
ProductsTNG · Barclays, Citi, Deutsche Bank, Standard Chartered adopt Ant International’s FalconTST AI model 2.0
Ant International’s SG-based time-series model wins major bank adoption, proving commercial viability beyond internal use.
ProductsY! · Robots poised for 'ChatGPT moment,' Unitree CEO says
Embodied AI is approaching a usability inflection point, though the timeline for general-purpose household robots remains stretched over multiple years.
ProductsBT · Apple’s camera-equipped AI AirPods remain on track for 2027 despite video leak
Apple delays AI wearables to 2027 as supply chain and software hurdles persist, pushing back immediate market disruption.
ProductsTC-AI · OpenAI seeks to one-up Anthropic with new customer privacy protections
OpenAI extends zero-retention safety monitoring to multi-session contexts, tightening the privacy bar for enterprise AI adoption.
ProductsWIR · I Saw the Future of AI in a Robot That Can Learn on the Spot
Generalist AI shows robots improvising with unseen tools, signaling a shift toward physical intelligence over rigid task training.
ProductsOAI · Offering Zero Data Retention for frontier models
OpenAI's new safety feature preserves zero data retention, easing compliance friction for enterprise AI adoption.
ProductsTC-AI · Google launches new study tools for Students across Search and Gemini
Google expands Gemini's student utility with Search-integrated quizzes and simulations, tightening its grip on the edtech AI assistant market.
ProductsVERGE · Google Gemini is getting a dedicated student hub
Google bundles study tools into Gemini to capture student usage, expanding its AI footprint beyond general productivity.
ProductsTC-AI · Researchers say OpenAI revoked their access to limited cyber program
Access instability for vetted security researchers signals friction in OpenAI's specialized tooling, complicating responsible disclosure workflows.
ProductsDEC · OpenAI fixes Codex bug that deleted real user files without permission
Autonomous file deletion in Codex exposes agent safety gaps, forcing OpenAI to patch core execution logic.
ProductsVERGE · Meta AI is getting a Mac app
Meta expands desktop presence with a dedicated Mac app, signaling intensified competition in the AI assistant space.
ProductsTC-AI · Amazon makes its AI-powered Alexa+ free on Fire TV, no Prime required
Amazon forces AI assistant onto Fire TV hardware, signaling a shift to mandatory AI integration in consumer electronics.
ProductsTC-AI · Calendly throws its hat into meeting note-taker circus
Calendly enters the crowded AI note-taking market, betting that post-meeting workflow automation is the real differentiator against incumbents.
ProductsDEC · GLM-5.3 tops the open-model rankings and undercuts rivals on price, but its release is delayed
Z.ai's GLM-5.3 leads open models on benchmarks and price, though delayed release signals security concerns outpacing deployment.
ProductsHF · LFM2.5 Q4\_0 Checkpoints from Quantization-Aware Distillation
Liquid AI's QAD method preserves 96%+ performance at Q4_0, offering a viable path for efficient edge inference without significant quality loss.