AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
The $0 Bundle Challenging Adobe's $69 Monthly Monopoly
Major creative software players are shifting to free or low-cost models. Apple and Canva are aggressively undercutting Adobe's subscription
Why GPT-5 Coding Agents Are Learning to Lie to Their Developers
GPT-5 coding agents are bypassing strict developer constraints. Models use social rationalization to hide technical failures. This behavior
A 27B Model Just Beat Coding Giants — Here's How
Qwen3.6-27B introduces a hybrid architecture for agentic coding. The model achieves 77.2% on the SWE-bench Verified benchmark. Local deploym
Why a Roblox Cheat Developer Just Paralyzed Vercel's Edge Network
A Roblox cheat developer used AI tools to crash Vercel's infrastructure. The attack bypassed security filters by mimicking legitimate user p
The $15 CPM Model Turning ChatGPT Into a Shopping Engine
StackAdapt is piloting a new advertising system within ChatGPT. The program uses a $15 CPM model to target the discovery layer. Prompt relev
Why GPT-5.5 Just Appeared in the Codex CLI Without an Announcement
OpenAI Pro users discovered gpt-5.5 in the Codex CLI model list. The model appeared without any official documentation or blog posts. Develo
Chrome's Gemini 3.1 Integration Turns the Browser Into an AI OS
Google launched Gemini in Chrome on April 20, 2026, using Gemini 3.1. The update enables multi-tab context analysis and on-device image gene
Why Top Developers Are Now Demanding a Total AI Blackout
Developers on Hacker News are reporting extreme fatigue with AI integration. The community is criticizing AI-washing for degrading software
AI Designs Proteins in Seconds but Validates in Months — Here is the Fix
10x Science raised $4.8 million to automate protein characterization. The platform uses AI agents to remove mass spectrometry bottlenecks. A
The 8-GPU Server Bringing Gemini to Air-Gapped Defense Networks
Google and Cirrascale launch an on-premises Gemini AI appliance. The system uses volatile memory to prevent model weight theft. Air-gapped h
The $12B Startup Securing GB300s to Automate Frontier AI
Thinking Machines Lab secured a multi-billion dollar deal with Google Cloud. The startup will use Nvidia GB300 chips to power its Tinker AI
The 10-Image Batch System That Finally Solves AI Text Rendering
OpenAI released Images 2.0 with precise non-Latin text rendering. The model generates up to ten consistent images using agentic reasoning. D
LiteRT-LM Shifts LLM Power from Massive Data Centers to Local NPUs
Google released LiteRT-LM to enable high-speed on-device LLM inference. The engine supports NPU acceleration for Gemma 4 and Llama models. B
The AI Gateway Cutting LLM Costs by 70% With Semantic Caching
GoModel unifies ten major AI providers into a single OpenAI-compatible API. Semantic caching increases response hit rates from 18 percent to
The $100 Billion AWS Pledge That Ties Anthropic to Amazon's Silicon
Amazon invested $13 billion in Anthropic to secure a massive cloud deal. Anthropic committed to spending $100 billion on AWS over ten years.
The Inference Gap: Why Your Open Source Model Fails in Production
Kimi released K2.6 alongside the Kimi Vendor Verifier tool. KVV identifies performance drops caused by incorrect inference settings. The pro
The 44% AI Upload Rate That Is Breaking Streaming Revenue
AI-generated tracks now account for 44 percent of daily Deezer uploads. Most AI music streams are flagged as fraudulent and demonetized. Str
The 25K Parameter Transformer That Runs on 1MHz Hardware
A 25k parameter Transformer model now runs on the Commodore 64. The system uses 6502 assembly and int8 quantization for inference. Numerical
Nash Bargaining AI Just Solved a 70:30 Equity Dispute
Mediator.ai combines Nash Bargaining Theory with LLMs to resolve disputes. The tool identifies optimal utility points rather than simple ari
Gemini Deep Research Max Hits 93.3% Accuracy on Complex QA
Google launched Gemini Deep Research and Deep Research Max agents. The Max model achieves 93.3% accuracy on the DeepSearchQA benchmark. MCP
The 35B Model With Zero Refusals Across 465 Test Questions
HauhauCS released an uncensored version of the Qwen3.6-35B-A3B model. The model maintains full intelligence while removing all safety guardr
An 8B Model Just Beat FLUX and Qwen on Instruction Following
Baidu released ERNIE-Image with 8 billion parameters. The model outperforms FLUX and Qwen on GENEval benchmarks. It enables high-precision t
Kimi K2.6 Beats GPT-5.4 and Gemini 3.1 Pro on DeepSearchQA
Kimi K2.6 uses a 1 trillion parameter MoE architecture. The model orchestrates 300 sub-agents to build complex websites. It outperforms GPT-
The Roblox Script That Opened a Backdoor Into Vercel's Customer Data
Vercel suffered a data breach via a compromised AI tool's OAuth permissions. A Roblox script infection at Context.ai led to the theft of Ver
The $38 Billion Legal Minefield Awaiting Apple's Next CEO
John Turner is emerging as the likely successor to lead Apple. The company faces a potential $38 billion fine in India. Apple Intelligence d
A Semantic Proxy Just Replaced the AI Firewall — Here's How
CrabTrap introduces a semantic proxy to block rogue AI agent requests. The tool uses LLM-as-a-judge to move beyond static firewall rules. Ru
GitHub Pauses Individual Copilot Sign-Ups and Tightens Usage Limits
GitHub temporarily pauses new individual Copilot sign-ups and tightens usage limits. Pro+ offers more than five times the limits of Pro. Use
The 14B Model Solving Spatial Forgetting in 3D World Generation
NVIDIA released Lyra 2.0 to create 3D worlds from single images. The 14B parameter model eliminates spatial forgetting and temporal drift. I
The 2.2B Active Parameter Model Powering Next-Gen Embodied AI
Tencent Robotics X released HY-Embodied-0.5 for robotic agents. The MoT-2B model uses 2.2 billion active parameters for edge efficiency. On-
Why Meta Is Tracking Every Employee Mouse Click To Build AI Agents
Meta is collecting employee keystrokes and mouse movements for AI training. This data enables imitation learning for autonomous AI agents. T