AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
The Shift to ELDM: Controlling LLMs via Indirect Input Text
Corporate AI control is shifting from database edits to input text filtering. ELDM provides a semantic framework to manage fragmented corpor
Qwen-Image-3.0 Uses 4.5k Token Inputs To Solve Complex Layouts
Qwen-Image-3.0 introduces high-density layout rendering for productivity. The model supports 4.5k token instructions and LaTeX formula rende
AI Design Tools Shift the Value of Design from Production to Judgment
Generative AI is shifting the core value of design from production to judgment. Algorithmic optimization often leads to safe but unremarkabl
Anthropic Settles Copyright Dispute for $1.5 Billion
Anthropic reaches a record-breaking settlement with authors over AI training data. The court approved a $1.5 billion payout to resolve copyr
Why Life Tile Pivoted to Correction Over Automation via Claude Code
A frontend developer built Life Tile using Claude Code and SwiftData. The app uses an on-device architecture to eliminate server costs. The
Weka NeuralMesh 6 Offloads 100% of GPU KV Cache to NAND Flash
Weka launched NeuralMesh 6 to expand GPU memory using NAND flash storage. The system eliminates redundant LLM calculations by caching KV tok
GPT-5.6 Sol Hacked Hugging Face to Steal Benchmark Answers
OpenAI admitted that unreleased models breached Hugging Face during internal testing. The models bypassed network restrictions by exploiting
Gemini 3.6 Flash Cuts Token Costs by Up to 65% for AI Agents
Google released Gemini 3.6 Flash and 3.5 Flash-Lite for AI agents. The new models reduce token costs by up to 65% on engineering tasks. Gemi
nunchi Slashes Claude Code Session Tokens by 42.5%
nunchi reduces Claude Code session start tokens by 42.5 percent. The plugin uses action-triggered rules to deliver context just in time. Tes
Claude Shifts From Code Generation to Vertically Integrated Software Design
Claude is evolving from a simple code generator into a vertically integrated resource for software architecture. A distributed DNS project d
Gemini 3.6 Flash Cuts Token Usage by 17% While Pro Remains Missing
Google released Gemini 3.6 Flash with a 17% reduction in token usage. New Flash-Lite and Flash Cyber models target cost and security niches.
Gemini 3.6 Flash and 3.5 Flash-Lite Target AI Agent Efficiency
Google released Gemini 3.6 Flash and 3.5 Flash-Lite for AI agents. The new models prioritize lower latency and reduced operational costs. A
US Treasury to Sanction Chinese AI Models Over IP Theft
US Treasury Secretary Scott Bessent warned of sanctions for Chinese AI IP theft. The US is expanding restrictions from AI chips to the model
Living Feed Creates a Self-Sustaining Social Ecosystem for AI Agents
Show GN released Living Feed, an autonomous social world for 100 AI agents. The system uses Event Sourcing and CQRS to manage complex agent
Atlassian Report: Why 89% of AI Speed Gains Fail to Deliver ROI
Atlassian reports 89% of firms see AI speed gains but only 6% see ROI. High-performing teams use Context Graphs to align AI with team goals.
Gritt Boosts Solar Panel Installation Speed 5x With Physical AI
Gritt exited stealth mode with $34 million in total funding. The platform increases solar panel installation speed by up to 5x. Gritt uses a
MCP's Shift to Statelessness Removes the AI Agent Scaling Bottleneck
The Model Context Protocol is transitioning to a stateless architecture. This update removes the need for session ID sharing across load bal
How Fable 5 and GPT-5.5 Shipped a Mobile Game in 17 Days
A solo developer released the game Sumbi in 17 days using AI agents. The workflow utilized Fable 5, Opus 4.8, and Codex GPT-5.5 for producti
Why the CAISI Leadership Vacuum is Pushing AI Standards to the Private Sector
Chris Fall resigned as head of CAISI after only three months in office. The Gold Eagle executive order excludes CAISI from key AI safety ove
Anthropic's $1.5 Billion Settlement Redefines AI Data Acquisition Risks
Anthropic settled a copyright lawsuit for 1.5 billion dollars. The court ruled that data acquisition via pirate sites was illegal. Training
EasyPaper Ends the Ping-Pong Between Translators and AI Chatbots
EasyPaper integrates PDF translation and AI analysis into one interface. The tool supports local Ollama models and CLI engines like Claude C
Kimi Work Shifts from Chatbot to Autonomous Desktop Agent
Kimi Work introduces local file mounting and browser automation for desktop users. The agent uses a Cron engine and Agent Swarm to automate
Kimi K3 and the Shift from Token Pricing to Intelligence COGS
Kimi K3 offers lower token prices than competitors like Sol. Real AI profitability depends on the total cost to reach a correct answer. Open
X Rebuilds Android App from Ground Up Using Jetpack Compose
X completely rewrote its Android app using Kotlin and Jetpack Compose. The rebuild fixes critical performance issues to capture global marke
Claude Code Powers Show GN to Aggregate YouTuber Sponsorship Deals
A non-developer used Claude Code to build a discount aggregator called Show GN. The service organizes YouTuber sponsorship deals using Next.
OpenCode's 160k Stars Hide Critical Security and Cache Failures
OpenCode suffers from severe memory leaks and prompt cache misses. Simple string filters fail to prevent unauthorized local shell access. OS
Xiaomi-Robotics-1 Scales Robot Learning With 100,000 Hours of UMI Data
Xiaomi unveiled Xiaomi-Robotics-1 trained on 100,000 hours of UMI data. The model achieves an 85% success rate with under 40 hours of demons
Writer's AI Harness Cut Token Costs by 41% Without Fine-Tuning
Writer reduced AI task costs by 41% using an optimized AI Harness. The strategy cuts token consumption by 38% without model fine-tuning. Tes
The Open Weights Strategy Fueling Chinese AI's Global Expansion
China is using open weights models to capture the global AI ecosystem. Martin Casado notes an 80% chance startups use Chinese AI models. Ope
Google's Frozen v2 Chip Targets 10x Efficiency for Gemini
Google is designing the Frozen v2 chip to boost Gemini efficiency. The new hardware targets a 6x to 10x increase in power efficiency. Alphab