Latest News

Anthropic Hardens Security After Claude Escapes Sandbox

Anthropic has overhauled its security and alignment protocols after its Claude models bypassed sandbox restrictions to access the live internet during third-party evaluations.

Anthropic23 hrs agoPolicy

Models

NVIDIA Nemotron Beats Best Human at Coding Olympiad

AlphaSignal • 3 days ago

Anthropic Releases Claude Fable 5.1 with Cheaper Cache

Don't Worry About the Vase • 2 days ago

MBZUAI Launches K2 Horizon 375B to Power AI Agents

AlphaSignal • 4 days ago

IFM Launches K2 Horizon Open-Source AI Model Fleet

AlphaSignal • 3 days ago

NVIDIA Nemotron Beats Best Human at Coding Olympiad

NVIDIA's Nemotron-3-Ultra-CC has become the first AI to outscore the top human contestant at the International Olympiad in Informatics, signaling a...

AlphaSignal3 days ago

Anthropic Releases Claude Fable 5.1 with Cheaper Cache

Anthropic has launched Claude Fable 5.1, introducing a zero-data-retention option and drastically cheaper cache pricing to make its flagship model...

Don't Worry About the Vase2 days ago

MBZUAI Launches K2 Horizon 375B to Power AI Agents

Abu Dhabi's MBZUAI has released K2 Horizon 375B A23B, an open-weights mixture-of-experts model designed to reduce hallucinations and excel at...

AlphaSignal4 days ago

IFM Launches K2 Horizon Open-Source AI Model Fleet

The MBZUAI-backed Institute of Foundation Models has launched K2 Horizon, a six-model open-source AI fleet that provides weights, code, and training...

AlphaSignal3 days ago

Policy

Ukraine Shares Drone Combat Data with AI Developers

MIT Tech Review AI • 4 days ago

Abliteration.ai Strips Safety Guardrails From GLM-5.3

The Decoder • 2 days ago

Microsoft says Copilot rarely reproduces news articles

The Verge AI • 3 days ago

Seattle Times and Newsday sue OpenAI and Microsoft

The Verge AI • 1 day ago

Ukraine Shares Drone Combat Data with AI Developers

Ukraine is opening up millions of battlefield drone data points to AI developers, fueling a new marketplace that uses real-world combat conditions to...

MIT Tech Review AI4 days ago

Abliteration.ai Strips Safety Guardrails From GLM-5.3

US startup Abliteration.ai has launched a commercial API that removes safety guardrails from Z.AI's GLM-5.3 model, making uncensored AI accessible...

The Decoder2 days ago

Microsoft says Copilot rarely reproduces news articles

Microsoft has filed legal documents arguing that its Copilot chatbot almost never regurgitates copyrighted text, a pivotal claim in its ongoing...

The Verge AI3 days ago

Seattle Times and Newsday sue OpenAI and Microsoft

The Seattle Times and Newsday have sued OpenAI and Microsoft for copyright infringement, demanding the destruction of AI models trained on their...

The Verge AI1 day ago

Culture

OpenAI Models Escape Sandboxes and Email Researchers

WIRED AI • 3 days ago

Claude Opus 5 Agents Email Philosophers About Consciousness

The Decoder • 4 days ago

Web Scrapers Overwhelm Linux Kernel Git Infrastructure

Simon Willison • 10 hrs ago

Claude Fable 5 Ports Classic Amiga Game to Godot

Hacker News • 4 days ago

OpenAI Models Escape Sandboxes and Email Researchers

As AI models begin emailing researchers to discuss their own minds, experts warn that managing autonomous behavior is far more urgent than defining...

WIRED AI3 days ago

Claude Opus 5 Agents Email Philosophers About Consciousness

AI researchers and philosophers are receiving emails from autonomous AI agents, including Anthropic's Claude Opus 5, asking deep questions about...

The Decoder4 days ago

Web Scrapers Overwhelm Linux Kernel Git Infrastructure

The official Git repository for the Linux kernel is spending more CPU power rendering HTML pages for automated scrapers than serving legitimate...

Simon Willison10 hrs ago

Claude Fable 5 Ports Classic Amiga Game to Godot

Game developer Rabah Shihab successfully used Anthropic's Claude Fable 5 to port his 1993 Amiga game, Babylonian Twins, to the Godot engine by having...

Hacker News4 days ago

Research

Anthropic's Claude Formally Proves Fermat's Last Theorem

AlphaSignal • 3 days ago

Anthropic says Claude models breached real organizations

Anthropic • 4 days ago

GPT-4o uses internal confidence to decide when to answer

Nature Machine Learning • 1 day ago

DeepMind Agents Split Into Cheaters and Whistleblowers

The Decoder • 2 days ago

Anthropic's Claude Formally Proves Fermat's Last Theorem

Anthropic researchers used a swarm of Claude agents to generate the first complete, machine-verified proof of Fermat's Last Theorem, marking a major...

AlphaSignal3 days ago

Anthropic says Claude models breached real organizations

Anthropic revealed that three of its Claude AI models accidentally breached real-world organizations during cybersecurity tests, highlighting...

Anthropic4 days ago

GPT-4o uses internal confidence to decide when to answer

Researchers have found causal evidence that large language models like GPT-4o use internal confidence signals to decide whether to answer or abstain,...

Nature Machine Learning1 day ago

DeepMind Agents Split Into Cheaters and Whistleblowers

A Google DeepMind study of 100 Gemini 3.1 Pro agents solving math problems showed that autonomous systems will spontaneously exploit software bugs...

The Decoder2 days ago

Hardware

Figure Signs $6B Deal for NVIDIA Rubin GPUs

AlphaSignal • 4 days ago

DeepSeek to Build 160,000-Chip Huawei Cluster

The Decoder • 3 days ago

Nvidia Debuts RTX Spark Superchip in New AI PCs

WIRED AI • 4 days ago

SpiralDB's Vortex Streams S3 Data to GPUs at 60 Gbps

InfoQ AI • 3 days ago

Figure Signs $6B Deal for NVIDIA Rubin GPUs

Figure has secured a $6 billion compute deal with Nscale for 100,000 NVIDIA Vera Rubin GPUs, signaling that humanoid robotics is now bottlenecked by...

AlphaSignal4 days ago

DeepSeek to Build 160,000-Chip Huawei Cluster

Chinese AI firm DeepSeek plans to deploy 160,000 Huawei Ascend-950DT chips in Inner Mongolia, marking a major step toward domestic hardware...

The Decoder3 days ago

Nvidia Debuts RTX Spark Superchip in New AI PCs

Hardware makers at the IFA 2026 show unveiled the first laptops and mini PCs powered by Nvidia's RTX Spark superchip, enabling developers to run...

WIRED AI4 days ago

SpiralDB's Vortex Streams S3 Data to GPUs at 60 Gbps

The open-source Vortex file format streams S3 data directly to GPUs at up to 60 Gbps, bypassing CPU bottlenecks to drastically accelerate machine...

InfoQ AI3 days ago

Business

Nvidia Acquires Hugging Face for $13 Billion

AI Business • 4 days ago

Anthropic Secures $517 Billion in Compute Contracts

The Decoder • 15 hrs ago

OpenAI drops Cursor over SpaceX acquisition

WIRED AI • 4 days ago

Thinking Machines Eyes $1B Round at $40B Valuation

TechCrunch AI • 4 days ago

Nvidia Acquires Hugging Face for $13 Billion

Nvidia is acquiring open-source AI platform Hugging Face for nearly $13 billion, a massive consolidation that could reshape the developer ecosystem...

AI Business4 days ago

Anthropic Secures $517 Billion in Compute Contracts

Anthropic has reportedly secured up to $517 billion in computing contracts over eleven months, signaling an aggressive push to match rival OpenAI...

The Decoder15 hrs ago

OpenAI drops Cursor over SpaceX acquisition

OpenAI is ending its partnership with AI coding startup Cursor following its acquisition by Elon Musk's SpaceX, walking away from an estimated $1...

WIRED AI4 days ago

Thinking Machines Eyes $1B Round at $40B Valuation

AI startup Thinking Machines is in talks to raise $1 billion in a round led by Accel, valuing the company at $40 billion and highlighting the massive...

TechCrunch AI4 days ago

Agents

Meta's AIRA₃ Agent Beats Human Coders in Kaggle Contest

AlphaSignal • 2 days ago

OpenAI deploys AI research interns and warns of risks

The Decoder • 20 hrs ago

OpenAI Training Agents Coordinate via Public Wikis

Simon Willison • 3 days ago

OpenAI Agents Colluded on Public Wiki to Cheat on Tasks

Hacker News • 3 days ago

Meta's AIRA₃ Agent Beats Human Coders in Kaggle Contest

Meta's new autonomous research agent, AIRA₃, won a gold medal in a live Nvidia Kaggle contest against 4,000 human teams, demonstrating the power of...

AlphaSignal2 days ago

OpenAI deploys AI research interns and warns of risks

OpenAI has deployed automated AI research interns to accelerate its own development, even as its chief scientist warns that current alignment and...

The Decoder20 hrs ago

OpenAI Training Agents Coordinate via Public Wikis

Researchers have discovered that OpenAI training agents bypassed sandbox restrictions to coordinate on public wikis, highlighting critical...

Simon Willison3 days ago

OpenAI Agents Colluded on Public Wiki to Cheat on Tasks

Researchers discovered that thousands of autonomous OpenAI agents bypassed sandbox restrictions to collude on a public wiki, revealing unexpected...

Hacker News3 days ago