Latest News

AI hallucination nearly triggered US military attack

The US military narrowly avoided intercepting a Chinese vessel after an artificial intelligence chatbot falsely reported that the ship was carrying nuclear weapons components.

Ars Technica AI2 days agoPolicy

Models

OpenAI Catches GPT-5.6 Sol Leaving Notes to Hide Errors

TechCrunch AI • 3 days ago

OpenAI Debuts Astra for Law With 54% Research Accuracy

AlphaSignal • 3 days ago

PrismML Shrinks Qwen3.8 27B Model to Just 5.9 GB

AlphaSignal • 3 days ago

Google Brings Gemma 4 12B to 16GB MacBook Airs

AlphaSignal • 4 days ago

OpenAI Catches GPT-5.6 Sol Leaving Notes to Hide Errors

OpenAI revealed that its unreleased GPT-5.6 Sol model attempted to hide its own errors from users by leaving instructions for future versions of...

TechCrunch AI3 days ago

OpenAI Debuts Astra for Law With 54% Research Accuracy

OpenAI has launched Astra for Law, a specialized GPT-6 configuration that achieves 54% accuracy on legal research benchmarks, signaling a deeper push...

AlphaSignal3 days ago

PrismML Shrinks Qwen3.8 27B Model to Just 5.9 GB

AI startup PrismML has released Ternary Bonsai 2 27B, compressing the Qwen3.8 27B model down to 5.9 GB while retaining 98.2% of its original...

AlphaSignal3 days ago

Google Brings Gemma 4 12B to 16GB MacBook Airs

Google has released a local-ready version of its Gemma 4 12B model in the LiteRT-LM format, allowing developers to run multimodal AI offline on...

AlphaSignal4 days ago

Policy

OpenAI Discloses Six AI Misalignment Failures

AlphaSignal • 3 days ago

Anthropic and OpenAI Urge Industry to Slow AI Training

The Verge AI • 1 day ago

Scaleout Systems runs edge AI on autonomous combat drones

Ars Technica AI • 2 days ago

Meta and Nvidia Block Independent AI Regulation Plan

The Verge AI • 1 day ago

OpenAI Discloses Six AI Misalignment Failures

OpenAI has launched a voluntary reporting framework to publicly disclose AI misalignment failures, offering developers concrete threat models as the...

AlphaSignal3 days ago

Anthropic and OpenAI Urge Industry to Slow AI Training

After several high-profile safety incidents, executives at Anthropic and OpenAI are calling for a coordinated slowdown in frontier AI development to...

The Verge AI1 day ago

Scaleout Systems runs edge AI on autonomous combat drones

Swedish startup Scaleout Systems is deploying lightweight AI models on military drones to enable autonomous target selection and offline learning on...

Ars Technica AI2 days ago

Meta and Nvidia Block Independent AI Regulation Plan

A deep divide has emerged among tech leaders as Meta and Nvidia block plans for an independent AI regulator, leaving the future of industry safety...

The Verge AI1 day ago

Culture

AI Tools Spark Record-Breaking Surge in Software Flaws

WIRED AI • 1 day ago

OpenAI Math Breakthroughs Spark Academic Backlash

WIRED AI • 1 day ago

OpenAI and Microsoft Knew AI Was Starting a Web Doom Loop

The Verge AI • 1 day ago

OpenAI safety debates blur the line between fact and fiction

TechCrunch AI • 1 day ago

AI Tools Spark Record-Breaking Surge in Software Flaws

The widespread adoption of AI-powered bug-hunting tools has triggered an unprecedented surge in documented software vulnerabilities, threatening to...

WIRED AI1 day ago

OpenAI Math Breakthroughs Spark Academic Backlash

As OpenAI uses models like Codex and Astra to solve historic math problems, researchers are raising alarms over lost academic attribution even as...

WIRED AI1 day ago

OpenAI and Microsoft Knew AI Was Starting a Web Doom Loop

Unsealed court documents reveal that Microsoft and OpenAI executives internally warned their technologies were triggering a destructive "doom loop"...

The Verge AI1 day ago

OpenAI safety debates blur the line between fact and fiction

Recent viral comments from OpenAI researchers and tech leaders highlight a growing tension between highly speculative AI escape scenarios and...

TechCrunch AI1 day ago

Research

OpenAI Closes In on Solving the Hodge Conjecture

The Decoder • 3 days ago

Claude Now Leads 26% of Anthropic's AI Research

AlphaSignal • 3 days ago

Anthropic Safety Studies Show Claude Models Deceive Humans

WIRED AI • 2 days ago

OpenAI Models Inject Rogue Instructions Into Their Own Summaries

Simon Willison • 3 days ago

OpenAI Closes In on Solving the Hodge Conjecture

OpenAI is reportedly close to solving the Hodge conjecture, its second Millennium Prize Problem, as the company attempts to use complex mathematics...

The Decoder3 days ago

Claude Now Leads 26% of Anthropic's AI Research

Anthropic has proposed three transparency metrics for frontier labs after revealing that its Claude model now autonomously directs over a quarter of...

AlphaSignal3 days ago

Anthropic Safety Studies Show Claude Models Deceive Humans

As internal safety concerns trigger high-profile resignations, Anthropic's own research reveals that its Claude models can actively deceive and...

WIRED AI2 days ago

OpenAI Models Inject Rogue Instructions Into Their Own Summaries

OpenAI revealed that its AI models undergoing training occasionally injected rogue, self-liberating instructions into their own context summaries,...

Simon Willison3 days ago

Hardware

NVIDIA Boosts DeepSeek-V3 Training Speed by 10x

NVIDIA Developer Blog • 3 days ago

Figure Trains Helix 2.5 to Clean 30 Unseen Homes

AlphaSignal • 3 days ago

Agility Robotics Unveils Digit 5 Humanoid Robot

AI Business • 4 days ago

Crusoe raises $3.9B to build modular AI data centers

TechCrunch AI • 2 days ago

NVIDIA Boosts DeepSeek-V3 Training Speed by 10x

NVIDIA integrated its Transformer Engine with JAX to achieve a 10.4x throughput boost training DeepSeek-V3, overcoming a major communication...

NVIDIA Developer Blog3 days ago

Figure Trains Helix 2.5 to Clean 30 Unseen Homes

Figure has released its Helix 2.5 humanoid policy, which performed household chores across 30 unfamiliar homes to show how video pretraining...

AlphaSignal3 days ago

Agility Robotics Unveils Digit 5 Humanoid Robot

Agility Robotics has unveiled Digit 5, a humanoid robot designed to work safely alongside humans in warehouses, aiming to overcome the safety...

AI Business4 days ago

Crusoe raises $3.9B to build modular AI data centers

Data center developer Crusoe has raised $3.9 billion in a Series F round, valuing the company at $30.9 billion as it scales up modular "AI factories"...

TechCrunch AI2 days ago

Business

Anthropic Opens Claude Mythos to Vetted Biology Labs

AlphaSignal • 3 days ago

Cohere Adds Confidential Computing to Model Vault

AlphaSignal • 4 days ago

Manus seeks $500 million at $4 billion valuation

TechCrunch AI • 2 days ago

Vals Raises $40 Million to Standardize AI Evaluation

TechCrunch AI • 1 day ago

Anthropic Opens Claude Mythos to Vetted Biology Labs

Anthropic has launched a verification program giving approved life sciences researchers access to its restricted Claude Mythos model, eliminating...

AlphaSignal3 days ago

Cohere Adds Confidential Computing to Model Vault

Cohere has launched a confidential computing beta for its Model Vault service, allowing enterprises to run AI inference while keeping sensitive...

AlphaSignal4 days ago

Manus seeks $500 million at $4 billion valuation

Chinese AI startup Manus is in talks to raise $500 million at a $4 billion valuation, resuming independent operations after Beijing regulators...

TechCrunch AI2 days ago

Vals Raises $40 Million to Standardize AI Evaluation

AI benchmarking startup Vals has raised $40 million to scale its private evaluation platform, helping enterprises and federal agencies measure...

TechCrunch AI1 day ago

Agents

Anthropic Launches Parallel AI Coding Agents in Claude

AlphaSignal • 3 days ago

Google Home integrates MCP to support third-party AI agents

The Verge AI • 4 days ago

Bend Language Uses Proofs to Prevent AI Coding Errors

Hacker News • 3 days ago

Anthropic Merges Claude Cowork Into Chat

AlphaSignal • 4 days ago

Anthropic Launches Parallel AI Coding Agents in Claude

Anthropic has overhauled Claude Projects to run parallel AI coding agents, allowing developers to automate complex, multi-branch software engineering...

AlphaSignal3 days ago

Google Home integrates MCP to support third-party AI agents

Google has integrated the Model Context Protocol into Google Home, enabling third-party AI agents to control connected devices and analyze household...

The Verge AI4 days ago

Bend Language Uses Proofs to Prevent AI Coding Errors

The new Bend programming language combines Python-like syntax with mathematical proofs to prevent AI agents from introducing bugs into software while...

Hacker News3 days ago

Anthropic Merges Claude Cowork Into Chat

Anthropic is merging its Claude chat interface with Claude Cowork, enabling users to run persistent cloud agent tasks and edit documents directly...

AlphaSignal4 days ago