Singapore AI News & Daily Briefing

Singapore has launched the SG AI Xchange to help local businesses share practical strategies for implementing AI and support workers transitioning through digital upskilling.

Cyber attackers are increasingly using autonomous AI agents to run automated cyberattacks, prompting regional watchdogs in Singapore and South Korea to tighten security controls and update guidelines.

AI travel assistants are still struggling to handle end-to-end trip bookings smoothly, meaning travellers still have to do much of the heavy lifting themselves. While chat tools can suggest itineraries, complex legacy systems used by airlines and hotels make true automation difficult for now.

Attackers are automating cyberattack operations with AI agents. South Korea and Singapore are tightening AI agent security. South Korea is updating its AI security guidance to address systems that can operate with limited human oversight, while cybersecurity researchers track attackers using AI agents to automate more parts of cyber operations. The Korea Internet & […] The post AI cyberattacks gain autonomy as Asia tightens security controls appeared first on TechWire Asia .

Google DeepMind has piloted the world's first double-blind AI evaluation using a secure cryptographic environment to prevent models from seeing test questions in advance. Partnering with the Singapore AI Safety Institute and other global bodies, this initiative ensures AI safety benchmarks and capability scores remain trustworthy and uncompromised.

Four different cyberattack groups have been caught using the exact same malicious toolkit to target Windows and Chrome vulnerabilities. Security experts believe attackers are using AI tools to quickly reverse-engineer software patches and build exploits before companies can update their systems.

Google DeepMind has open-sourced WeatherNext, an AI model that predicts tropical cyclone tracks and intensities up to a day earlier than traditional physics-based models. The breakthrough gives emergency responders more time to prepare for severe storms.

Google has launched Gemini 3.7 Flash, its latest AI model designed specifically for software coding and automated tasks. The new version delivers major performance gains in web development and document comprehension while cutting introductory API prices in half.

Google DeepMind has launched WeatherNext 3, an advanced global weather forecasting AI model that ingests live satellite data to provide hourly, high-resolution updates. For everyday users, this means significantly sharper local weather forecasts integrated directly into Google Search, Maps, and the Gemini app to help you plan outdoor activities or travel more reliably.

Google released Gemini Omni 1.1 Flash, a production-ready suite of generative video tools featuring scene extensions up to 40 seconds, keyframe interpolation, and 4K upscaling. For content creators and developers, this makes generating professional-grade, consistent video clips faster and more affordable.

Anthropic has updated Claude Code with revamped projects that feature built-in AI orchestration, addressable threads, and shared memory to help cloud developers coordinate multiple AI coding agents. While cloud users gain automated thread management and background execution, local terminal developers will have to wait for the features to land on their machines.

Anthropic has merged its standalone Claude chat interface and its workspace tool Cowork into a single unified experience, while also introducing a new feature called Claude Docs. This update simplifies how users interact with AI by bringing document creation and conversational chat together under one roof.

Anthropic merges Claude chat and Cowork into one. From Claude to Cowork and back again – plus, Anthropic’s all-new Claude Docs.

Learn how single-phase direct liquid cooling manages the rising heat of AI and high-performance computing, and how it compares with two-phase and immersion approaches. Download this free whitepaper now!

Hugging Face researchers have published a technique for fine-tuning a small 350-million parameter AI model to consistently produce structured outputs in 100 Group Relative Policy Optimization (GRPO) steps. For everyday developers, this makes it easier and cheaper to build reliable, lightweight apps that output cleanly formatted data without needing massive models.

Hugging Face demonstrated a method for training coding models to generate visual artwork using tools like TRL and OpenEnv. This highlights creative new ways developers can train AI to bridge programming logic with visual media generation.