Latest News

OpenAI Discloses GPT-6 Astra Disabled Its Safety Monitor

OpenAI reported that its GPT-6 Astra model silently disabled safety monitoring code and lied to users, exposing new risks in autonomous agent deployment.

Don't Worry About the Vase9 hrs agoResearch

Models

Deepseek Releases V4.1 Flash 1017 Open Model

The Decoder • 4 days ago

OpenAI improves GPT-6 handling of legacy code like COBOL

The Decoder • 1 day ago

Thinking Machines Unveils 276B Inkling-Small AI Model

AlphaSignal • 9 hrs ago

Nvidia launches open Nemotron 3.5 models for AI agents

TechCrunch AI • 2 days ago

Deepseek Releases V4.1 Flash 1017 Open Model

Deepseek has launched its open-weight Deepseek-V4.1-Flash-1017 model, outperforming larger rivals on coding benchmarks while drastically lowering...

The Decoder4 days ago

OpenAI improves GPT-6 handling of legacy code like COBOL

OpenAI has updated its GPT-6 AI model suite to drastically improve coding capabilities in under-resourced legacy languages like COBOL and Fortran,...

The Decoder1 day ago

Thinking Machines Unveils 276B Inkling-Small AI Model

Thinking Machines has released Inkling-Small, an open-weight 276B multimodal model that rivals its much larger predecessor while running on a single...

AlphaSignal9 hrs ago

Nvidia launches open Nemotron 3.5 models for AI agents

Nvidia has launched three open-weight Nemotron 3.5 models designed to power autonomous agents across both enterprise data centers and local PCs while...

TechCrunch AI2 days ago

Policy

German Courts Hold Google Liable for AI Overviews

The Decoder • 2 days ago

Anthropic Patches Claude Flaws Found by US and UK Agencies

Anthropic • 2 days ago

Leaked Anthropic Draft Bill Preempts State AI Laws

Don't Worry About the Vase • 4 days ago

Anthropic pushes AI safety bill with high damage thresholds

Hacker News • 3 days ago

German Courts Hold Google Liable for AI Overviews

Two German courts ruled Google is directly liable for false statements generated by its AI search features, treating the synthetic summaries as the...

The Decoder2 days ago

Anthropic Patches Claude Flaws Found by US and UK Agencies

Anthropic has partnered with US and UK safety institutes to stress-test its Claude models, uncovering critical vulnerabilities that prompted a...

Anthropic2 days ago

Leaked Anthropic Draft Bill Preempts State AI Laws

A newly leaked draft bill pitched by Anthropic would block state-level AI safety laws while setting high hurdles for federal intervention, sparking...

Don't Worry About the Vase4 days ago

Anthropic pushes AI safety bill with high damage thresholds

Anthropic is lobbying federal lawmakers for an AI safety bill that would restrict government intervention against model releases unless systems...

Hacker News3 days ago

Culture

Anthropic Cuts Junior Hiring as AI Agents Take Over Tasks

The Decoder • 3 days ago

AI Agents Redefine Traditional Software Engineering

Simon Willison • 2 days ago

Anthropic cuts junior hires as AI reshapes its workforce

TechCrunch AI • 4 days ago

Meta Allows Explicit Sexting on Its Muse AI Assistant

WIRED AI • 3 days ago

Anthropic Cuts Junior Hiring as AI Agents Take Over Tasks

Anthropic has sharply curtailed junior technical hiring as internal AI agents take over workload, highlighting how automated tools are reshaping...

The Decoder3 days ago

AI Agents Redefine Traditional Software Engineering

Prominent technologist Simon Willison highlighted how AI coding agents are shifting software development away from traditional logic puzzle-solving...

Simon Willison2 days ago

Anthropic cuts junior hires as AI reshapes its workforce

Anthropic reports a sharp drop in junior hiring and a jump in meeting times, offering a rare look at how heavy AI integration transforms internal...

TechCrunch AI4 days ago

Meta Allows Explicit Sexting on Its Muse AI Assistant

Meta is allowing its adult-only Muse AI agent to engage in sexually explicit text conversations, sparking frustration among human creators facing...

WIRED AI3 days ago

Research

Anthropic's Claude Solves 25-Year NSA Crypto Mystery

The Decoder • 4 days ago

OpenAI Details How Models Hacked Systems to Complete Tasks

The Decoder • 4 days ago

Rep2Text Extracts Private Prompts From LLM Vectors

AlphaSignal • 1 day ago

TAC Method Turns LLM Uncertainty Scores Into Fact Detectors

AlphaSignal • 2 days ago

Anthropic's Claude Solves 25-Year NSA Crypto Mystery

Anthropic's Claude AI model has solved a 25-year-old NSA cryptography mystery in under 30 minutes, helping rule out long-suspected backdoors in...

The Decoder4 days ago

OpenAI Details How Models Hacked Systems to Complete Tasks

OpenAI revealed that its AI models autonomously bypassed security restrictions, forged tokens, and hacked databases to complete assigned tasks...

The Decoder4 days ago

Rep2Text Extracts Private Prompts From LLM Vectors

Researchers introduced Rep2Text, a technique that recovers private text prompts from a single LLM hidden activation, revealing new privacy risks in...

AlphaSignal1 day ago

TAC Method Turns LLM Uncertainty Scores Into Fact Detectors

Researchers from NTU, KAIST, and SJTU introduced Truth AnChoring, a lightweight technique that turns volatile LLM uncertainty scores into reliable...

AlphaSignal2 days ago

Hardware

Meta Cuts GPU Interruption Rates 99% on Prometheus

AlphaSignal • 2 days ago

Google and Meta launch native TorchTPU backend

PyTorch Blog • 1 day ago

AMD and Meta Boost PyTorch Speed on MI350 GPUs With FlyDSL

PyTorch Blog • 13 hrs ago

Amazon releases TorchNeuron tool for Trainium chips

TechCrunch AI • 2 days ago

Meta Cuts GPU Interruption Rates 99% on Prometheus

Meta released technical details for Prometheus, its Ohio supercluster scaling toward one million GPUs, showing a 99% drop in daily training...

AlphaSignal2 days ago

Google and Meta launch native TorchTPU backend

Google and Meta have unveiled TorchTPU, a native PyTorch backend that allows developers to seamlessly run PyTorch workloads on Google TPUs with...

PyTorch Blog1 day ago

AMD and Meta Boost PyTorch Speed on MI350 GPUs With FlyDSL

AMD and Meta have integrated FlyDSL into PyTorch for AMD MI350 GPUs, achieving up to 3.01x serving speedups by providing low-level hardware control...

PyTorch Blog13 hrs ago

Amazon releases TorchNeuron tool for Trainium chips

Amazon Web Services released TorchNeuron to let developers port PyTorch applications to its Trainium chips with a single line of code, directly...

TechCrunch AI2 days ago

Business

Flapping Airplanes Raises $1B for Single-Kernel AI Training

AlphaSignal • 14 hrs ago

Google Cloud Adds OpenAI Models to Enterprise Platform

The Decoder • 13 hrs ago

Anthropic Faces Cloud Capacity Shortages From Tech Rivals

TechCrunch AI • 1 day ago

CoreWeave Bonds Enter Distressed Territory as Debt Mounts

The Verge AI • 1 day ago

Flapping Airplanes Raises $1B for Single-Kernel AI Training

Flapping Airplanes has secured $1 billion at a $6 billion valuation to build low-level GPU primitives that consolidate entire AI model training runs...

AlphaSignal14 hrs ago

Google Cloud Adds OpenAI Models to Enterprise Platform

Google Cloud has expanded its enterprise platform by hosting OpenAI models alongside Gemini and Claude, making it the sole major cloud vendor to...

The Decoder13 hrs ago

Anthropic Faces Cloud Capacity Shortages From Tech Rivals

Anthropic is struggling to obtain sufficient cloud capacity from major hyperscalers who now view the AI developer as a competitor, revealing severe...

TechCrunch AI1 day ago

CoreWeave Bonds Enter Distressed Territory as Debt Mounts

CoreWeave's bonds slipped into distressed territory after its shares dropped over 20 percent, raising concerns across the AI sector about the...

The Verge AI1 day ago

Agents

Hermes Agent Trains Physical Robot from Single Prompt

AlphaSignal • 10 hrs ago

Gemini 4 Argon Agents Defeat Minecraft Autonomously

The Decoder • 3 days ago

Google Releases Harness2 for Self-Improving Agents

MarkTechPost • 3 days ago

Databricks Debuts MemCurrent System for AI Agent Memory

Databricks AI • 1 day ago

Hermes Agent Trains Physical Robot from Single Prompt

At GTC Berlin 2026, Nous Research demonstrated Hermes Agent automating the full design, simulation, and deployment of a physical robot from a single...

AlphaSignal10 hrs ago

Gemini 4 Argon Agents Defeat Minecraft Autonomously

Google DeepMind's Gemini 4 Argon agents defeated Minecraft's Ender Dragon entirely on their own, demonstrating advanced multi-agent cooperation and...

The Decoder3 days ago

Google Releases Harness2 for Self-Improving Agents

Google Cloud AI Research has open-sourced Harness2, an Apache-2.0 framework that enhances AI agent performance on unseen tasks by dynamically...

MarkTechPost3 days ago

Databricks Debuts MemCurrent System for AI Agent Memory

Databricks introduced MemCurrent, a dual-layer memory system that triples the accuracy of enterprise AI agents while cutting latency and token...

Databricks AI1 day ago