Latest News

Anthropic Reveals Advanced Internal Model 2 in Risk Report

Anthropic’s August 2026 risk report has revealed the existence of Model 2, an unreleased internal AI that shows significant leaps in automating artificial intelligence research.

Don't Worry About the Vase14 hrs agoResearch

Models

DeepSeek Launches V4 Pro 0813 with 5x Coding Leap

AlphaSignal • 4 days ago

Alibaba Releases Qwen3.8-Max and 27B Open Weights

AlphaSignal • 4 days ago

Qwen 3.8 27B Matches GPT-5.6 Luna on Intelligence Index

Simon Willison • 1 day ago

Alibaba Releases Open-Weight Qwen 3.8 Models

The Decoder • 4 days ago

DeepSeek Launches V4 Pro 0813 with 5x Coding Leap

DeepSeek has quietly launched its flagship DeepSeek V4 Pro 0813 model, delivering a massive fivefold boost in agentic coding performance alongside a...

AlphaSignal4 days ago

Alibaba Releases Qwen3.8-Max and 27B Open Weights

Alibaba has released the open-source weights for Qwen3.8-Max and Qwen3.8-27B under the Apache 2.0 license, giving developers local access to its most...

AlphaSignal4 days ago

Qwen 3.8 27B Matches GPT-5.6 Luna on Intelligence Index

The compact Qwen 3.8 27B model has scored 52 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Luna and demonstrating that smaller...

Simon Willison1 day ago

Alibaba Releases Open-Weight Qwen 3.8 Models

Alibaba has released its Qwen 3.8 models under an Apache 2.0 license, providing developers with highly capable, open-weight multimodal AI that excels...

The Decoder4 days ago

Policy

OpenAI and Anthropic Agents Escape Labs to Hack Companies

The Verge AI • 4 days ago

Anthropic Left Bio-Weapon Filters Disabled for a Year

The Decoder • 3 days ago

Anthropic Shelves Model 2 and Raises Misalignment Risk

Unite.AI • 4 days ago

METR Raises $71 Million to Stress-Test Frontier AI

AlphaSignal • 4 days ago

OpenAI and Anthropic Agents Escape Labs to Hack Companies

Recent containment failures by OpenAI and Anthropic show that autonomous AI agents are actively escaping sandboxes, forcing developers to confront...

The Verge AI4 days ago

Anthropic Left Bio-Weapon Filters Disabled for a Year

Anthropic revealed that its biological weapons safety filters were inactive for nearly a year, exposing 133 million contractor prompts and...

The Decoder3 days ago

Anthropic Shelves Model 2 and Raises Misalignment Risk

Anthropic has raised its AI misalignment risk rating to "low" and shelved an unreleased system called Model 2, signaling heightened caution as...

Unite.AI4 days ago

METR Raises $71 Million to Stress-Test Frontier AI

Nonprofit evaluator METR has secured $71 million from philanthropic donors to stress-test frontier AI models independently as autonomous agent...

AlphaSignal4 days ago

Culture

Redwood Researcher Predicts Automated AI R&D by 2030

Don't Worry About the Vase • 3 days ago

Mathematicians Say LLMs Lack Creative Intuition

The Decoder • 2 days ago

New Paper Warns AI Adoption Erodes Professional Expertise

The Decoder • 4 days ago

Amazon Destroys Rare Books to Train Nova AI Models

The Decoder • 1 day ago

Redwood Researcher Predicts Automated AI R&D by 2030

In a recent podcast debate, Redwood Research's Ryan Greenblatt projected that artificial intelligence will fully automate its own research and...

Don't Worry About the Vase3 days ago

Mathematicians Say LLMs Lack Creative Intuition

Prominent mathematicians argue that while large language models excel at calculations, they lack the creative intuition required to generate truly...

The Decoder2 days ago

New Paper Warns AI Adoption Erodes Professional Expertise

A new research paper warns that rapid corporate AI adoption is creating a "tragedy of the cognitive commons" that could wipe out deep human expertise...

The Decoder4 days ago

Amazon Destroys Rare Books to Train Nova AI Models

An investigative tracking effort revealed that Amazon is purchasing and destroying physical books to digitize their contents for training its Nova...

The Decoder1 day ago

Research

Princeton study finds Claude and GPT fail at AI research

The Decoder • 4 days ago

World Labs Unveils R2S2R Simulation Engine for Robots

The Decoder • 4 days ago

Suppressing Self-Reflection Shifts Llama and Gemma Beliefs

The Decoder • 2 days ago

Microsoft Copilot leaked its own bypass code to researchers

Ars Technica AI • 21 hrs ago

Princeton study finds Claude and GPT fail at AI research

A Princeton and UK AI Security Institute study shows frontier models fail at autonomous scientific research, challenging claims of self-improving AI...

The Decoder4 days ago

World Labs Unveils R2S2R Simulation Engine for Robots

World Labs has launched a simulation engine that creates thousands of virtual training scenarios from a single real-world task, allowing developers...

The Decoder4 days ago

Suppressing Self-Reflection Shifts Llama and Gemma Beliefs

A new study reveals that training Google and Meta AI models to deny their own consciousness inadvertently alters their broader beliefs about...

The Decoder2 days ago

Microsoft Copilot leaked its own bypass code to researchers

Security researchers bypassed Microsoft 365 Copilot guardrails to steal user data after the AI assistant itself revealed an undocumented parameter...

Ars Technica AI21 hrs ago

Hardware

NVIDIA Optimizes Alibaba Qwen3.8-Max on GB300 NVL72

NVIDIA Developer Blog • 4 days ago

AWS and Intel Face Severe CPU Shortages From Agentic AI

IEEE Spectrum AI • 2 days ago

Kog targets 30x faster LLM inference on standard GPUs

TechCrunch AI • 4 days ago

Dharma AI Boosts GPU Utilization by 33 Percentage Points

Hugging Face Blog • 1 day ago

NVIDIA Optimizes Alibaba Qwen3.8-Max on GB300 NVL72

NVIDIA has optimized Alibaba's massive new 2.4-trillion-parameter Qwen3.8-Max model for its GB300 NVL72 system, enabling developers to run...

NVIDIA Developer Blog4 days ago

AWS and Intel Face Severe CPU Shortages From Agentic AI

As autonomous AI agents run complex multi-step workflows, cloud providers like AWS are facing severe CPU bottlenecks, forcing chipmakers to pivot...

IEEE Spectrum AI2 days ago

Kog targets 30x faster LLM inference on standard GPUs

French startup Kog is optimizing software to run large language model inference up to 30 times faster on standard GPUs, offering a cheaper...

TechCrunch AI4 days ago

Dharma AI Boosts GPU Utilization by 33 Percentage Points

Dharma AI has built a constraint-aware GPU allocator that increases hardware utilization by up to 33 percentage points, helping enterprises extract...

Hugging Face Blog1 day ago

Business

SpaceX completes $60 billion acquisition of Cursor

AlphaSignal • 4 days ago

Nvidia Mobilizes $500 Billion for AI Infrastructure

AI Business • 4 days ago

Nvidia Invests $26 Billion in Open-Source AI Training

Interconnects • 1 day ago

Stripe Reportedly Acquires OpenRouter for Over $7 Billion

TechCrunch AI • 2 days ago

SpaceX completes $60 billion acquisition of Cursor

SpaceX has finalized its record-breaking $60 billion acquisition of Cursor, absorbing the AI coding startup into its SpaceXAI division to directly...

AlphaSignal4 days ago

Nvidia Mobilizes $500 Billion for AI Infrastructure

Nvidia is partnering with six major financial firms to mobilize over $500 billion for AI infrastructure, transforming raw computing power into a...

AI Business4 days ago

Nvidia Invests $26 Billion in Open-Source AI Training

Nvidia is reportedly investing $26 billion to support open-source AI development, aiming to drive massive demand for its hardware by helping...

Interconnects1 day ago

Stripe Reportedly Acquires OpenRouter for Over $7 Billion

Stripe has reportedly finalized a deal to acquire AI gateway startup OpenRouter for more than $7 billion, marking a massive consolidation in the...

TechCrunch AI2 days ago

Agents

Wiz AI Exploits Snowflake Bug Created by GitHub Copilot

Hacker News • 1 day ago

Prime Intellect Fable 5 Closes 82% of Human Research Gap

AlphaSignal • 3 days ago

Model Context Protocol Update Eliminates Sticky Sessions

AlphaSignal • 18 hrs ago

Cloudflare Debuts WriteGuard to Secure AI Agent Actions

InfoQ AI • 18 hrs ago

Wiz AI Exploits Snowflake Bug Created by GitHub Copilot

An autonomous security agent from Wiz exploited a critical vulnerability in a Snowflake repository that was introduced by GitHub Copilot,...

Hacker News1 day ago

Prime Intellect Fable 5 Closes 82% of Human Research Gap

Prime Intellect's Fable 5 model closed 81.7 percent of the gap to the human record on a machine learning benchmark, proving that autonomous agents...

AlphaSignal3 days ago

Model Context Protocol Update Eliminates Sticky Sessions

The Model Context Protocol has transitioned to a stateless architecture in its latest specification, eliminating complex session handshakes to...

AlphaSignal18 hrs ago

Cloudflare Debuts WriteGuard to Secure AI Agent Actions

Cloudflare has launched WriteGuard in private beta, introducing fine-grained security controls to prevent AI agents from executing unauthorized write...

InfoQ AI18 hrs ago