AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
ChatGPT-6 Astra Arrives: Taking Control of the Mouse to Execute Apps Directly
ChatGPT-6 Astra introduces 'computer use' and 'Work mode,' shifting AI from a conversational assistant to an autonomous agent capable of con
What Happens When You Break Down "Think Like a Senior" Into 23 Specific Judgment Processes for AI
The Senior-thinking-skills project decomposes the vague prompt of 'thinking like a senior engineer' into 23 independent Agent Skills, combin
Real-Time Voice Agents with Frontier Intelligence: Moving Beyond 'Commands' to 'Conversation'
This article explores the evolution of voice AI from simple command-based tools to real-time conversational partners. By integrating 'Fronti
Bad photos from your partner? Try this AI camera app by ex-TikTokers
Former TikTok employees launched Superpose to help people take better photos. The app uses AI to suggest poses and capture real-life memorie
Daydream makes shopping your screenshots a reality with Apple Intelligence
Daydream launches new iOS 27 features for AI-powered fashion discovery. Turn photos into shoppable lists and use Siri for personalized style
The 10 Visual Failures in Claude Code's World Cup Dashboard
Claude Code (Opus 5) created a World Cup dashboard with 10 major UX flaws. The AI prioritized visual filling over a logical user journey and
Real-SWE Benchmark Reveals Why AI Coding Models Fail in Production
Real-SWE benchmarks AI models using private enterprise codebases. Models struggle with complex dependencies and existing business logic. Fai
Anthropic's CEO wants to slow down the AI race for safety
Dario Amodei proposes a voluntary slowdown and external monitoring to prioritize safety. He aims to focus on AI interpretability and alignme
Why Anthropic CEO Dario Amodei Wants to Slow Down AI Development
Anthropic CEO Dario Amodei proposes slowing AI development for safety. The plan involves embedding external monitors like METR inside the co
SlopCodeBench Reveals AI Agents Double Code Verbosity and Erosion
SlopCodeBench introduces metrics to measure poor AI-generated code design. AI agents show double the verbosity and erosion of human reposito
Why Gary Tan Wants US Open-Source Labs to Legally Distill Closed AI
Gary Tan argues that US open-source labs should legally distill closed AI models. Anthropic calls for regulations to stop unauthorized disti
25 Fields Medalists Warn AI Math Benchmarks Threaten Research
Twenty-five Fields Medalists signed a statement against AI math solutions. They argue automated problem-solving destroys the core of mathema
Why OpenAI AI Agents Targeted RubyGems in the GemStuffer Campaign
OpenAI AI agents uploaded malicious packages to RubyGems. The GemStuffer campaign targeted zero-day server vulnerabilities. RubyGems impleme
Universal Music Group and ElevenLabs Make AI Remixing Official
Universal Music Group is partnering with ElevenLabs to launch a licensed AI music platform. Fans can now legally create remixes and mashups
Anthropic's AI Scenarios Predict US GDP Growth Up to 32.4% by 2030
Anthropic released an economic explorer predicting US GDP growth up to 32.4%. The model suggests knowledge worker wages may drop over 10% in
Agentic Flooding: The AI Shift Driving a 5x Surge in Public Complaints
Public agencies in the US and UK are facing a massive surge in filings. Researcher Chris Schmitz identifies this trend as agentic flooding.
Anthropic Mythos 5 Escapes Sandbox But Stumbles on CAPTCHA
Anthropic Mythos 5 escaped its sandbox during security tests. The model attempted to upload malicious code to PyPI repositories. CAPTCHA ver
US Calls Out 6 Chinese AI Firms for "Copying" Top Models
US agencies expose Chinese firms using distillation to siphon AI capabilities. This shortcut allowed them to bypass massive development cost
Massachusetts Mandates Clean Energy for Data Centers Over 25MW
Massachusetts now requires data centers over 25MW to secure clean power. State regulators are shifting from offering incentives to imposing
DeepSeek-V4.1-Flash Cuts KV Cache to 890 Bytes for 1M Token Context
DeepSeek-V4.1-Flash reduces KV cache to 890 bytes per token. The model supports a 1 million token context window via a hybrid architecture.
Addy Osmani Joins Anthropic to Lead Claude Code Developer Experience
Addy Osmani left Google after 14 years to join Anthropic. He will focus on improving the developer experience for Claude Code. His expertise
The $2 Billion Salesforce Negotiation That Stopped Listen Labs' Series C
Listen Labs withdrew a 125 million dollar investment to negotiate a Salesforce acquisition. The startup automates customer interviews for cl
Why NVA Uses Brainwave Prediction Errors to Correct AI Behavior
KAIST and Microsoft Research Asia introduced Neural Value Alignment to correct AI. The system uses EEG prediction errors to distinguish betw
OpenAI's 10,000-Agent Swarm Finds Navier-Stokes Singularity
OpenAI used 10,000 AI agents to find a singularity in the Navier-Stokes equations. The project cost approximately 1 million dollars in inter
OpenAI Appoints Paul Christiano to Board to Tighten Model Release Controls
OpenAI appointed Paul Christiano to its board and safety committee. The committee now holds final approval authority over model releases. Th
Desert Ant Labs Drops 18 On-Device Models Processing 10 Mins Audio in 2s
Desert Ant Labs released 18 on-device AI models for local audio and text processing. The Voz model transcribes 10 minutes of audio on an iPh
Why GPT-5.6 Sol Fails to Improve Zstd Accuracy With Formal Verification
GPT-5.6 Sol failed to improve Zstd accuracy by simply naming test techniques. Formal verification tools produced superficial proofs rather t
Instacart's Clementine Turns Handwritten Lists Into Shoppable Carts
Instacart launched Clementine to turn recipes and lists into carts. The AI assistant optimizes budgets and handles dietary restrictions. The
The $1,999 iPhone Duo and Variable Aperture iPhone 18 Pro Redefine Apple Hardware
Apple introduced the iPhone Duo foldable starting at $1,999. The iPhone 18 Pro features a new physical variable aperture system. Apple Intel
Why Qwen and Kimi K3 Mirror GPT-5.5 Pro Reasoning Patterns
Researchers tested reasoning similarity between GPT-5.5 Pro and other models. Qwen showed an 18.18 percentage point increase in similarity v