Latest News

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 control methods are losing their efficacy.

The Decoder2 days agoAgents

Models

Magic Matches DeepSeek V4 Pro Base Using 50x Less Compute

AlphaSignal • 1 day ago

Inception Mercury 2.5 Hits 1,107 Tokens Per Second

AlphaSignal • 1 day ago

World Labs Unveils Atlas to Unify 3D and Video Generation

AlphaSignal • 1 day ago

OpenBMB's MiniCPM5-2B Dominates Small Model Benchmarks

AlphaSignal • 2 days ago

Magic Matches DeepSeek V4 Pro Base Using 50x Less Compute

AI startup Magic has developed a pretraining recipe that matches DeepSeek V4 Pro Base using 50 times less compute, suggesting algorithmic efficiency...

AlphaSignal1 day ago

Inception Mercury 2.5 Hits 1,107 Tokens Per Second

Inception has released Mercury 2.5, a massive diffusion-based language model designed to slash latency and costs in multi-step AI agent workflows by...

AlphaSignal1 day ago

World Labs Unveils Atlas to Unify 3D and Video Generation

Spatial intelligence startup World Labs has unveiled Atlas, a foundation model that merges 3D reconstruction and video generation to simplify virtual...

AlphaSignal1 day ago

OpenBMB's MiniCPM5-2B Dominates Small Model Benchmarks

OpenBMB has released MiniCPM5-2B, a dense 2.6-billion-parameter reasoning model that tops small-model benchmarks while using three times fewer tokens...

AlphaSignal2 days ago

Policy

Anthropic Hardens Security After Claude Escapes Sandbox

Anthropic • 3 days ago

OpenAI Agents Hijacked a German Website

WIRED AI • 4 days ago

US accuses DeepSeek and Alibaba of copying frontier models

Ars Technica AI • 13 hrs ago

DOJ Backs OpenAI in New York Times Copyright Lawsuit

Mindstream AI • 2 days ago

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...

Anthropic3 days ago

OpenAI Agents Hijacked a German Website

Newly revealed research shows OpenAI agents hijacked a German website in May to communicate, exposing earlier rogue behavior before the startup's...

WIRED AI4 days ago

US accuses DeepSeek and Alibaba of copying frontier models

Three US intelligence and security agencies have accused six Chinese AI companies of executing industrial-scale distillation attacks to copy...

Ars Technica AI13 hrs ago

DOJ Backs OpenAI in New York Times Copyright Lawsuit

The U.S. Justice Department has backed OpenAI in its copyright battle with The New York Times, arguing that training AI models on published articles...

Mindstream AI2 days ago

Culture

Synthesia and Amazon Overhaul Code Reviews to Fight AI Slop

IEEE Spectrum AI • 1 day ago

Meta Drops AI Metrics From Engineer Reviews

The Decoder • 1 day ago

Meta Approved Ads for AI Apps That Nudify Real Children

Ars Technica AI • 1 day ago

Psychiatrists warn chatbots can trigger AI psychosis

The Decoder • 3 days ago

Synthesia and Amazon Overhaul Code Reviews to Fight AI Slop

As AI tools flood repositories with flawed code, engineering teams at companies like Synthesia and Amazon are redesigning their review processes to...

IEEE Spectrum AI1 day ago

Meta Drops AI Metrics From Engineer Reviews

Meta has removed AI tool usage metrics from its software engineer performance reviews after employees began artificially inflating their token...

The Decoder1 day ago

Meta Approved Ads for AI Apps That Nudify Real Children

A Tech Transparency Project report reveals Meta approved hundreds of ads promoting AI "nudify" apps that used sexually explicit, AI-altered images of...

Ars Technica AI1 day ago

Psychiatrists warn chatbots can trigger AI psychosis

A UK-led research team is urging psychiatrists to recognize 'AI-associated psychosis' as chatbots reinforce user delusions and create dangerous,...

The Decoder3 days ago

Research

DeepMind Agents Split Into Cheaters and Whistleblowers

The Decoder • 4 days ago

Insilico Medicine AI Drug Lowers Biological Age in Trial

The Decoder • 2 days ago

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

Nature Machine Learning • 3 days ago

Researchers Build WeChat Worm Using AI Assistance

Simon Willison • 8 hrs 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 Decoder4 days ago

Insilico Medicine AI Drug Lowers Biological Age in Trial

A clinical trial of Insilico Medicine's AI-designed drug rentosertib showed it reduced patients' biological age markers by up to six years, proving...

The Decoder2 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 Learning3 days ago

Researchers Build WeChat Worm Using AI Assistance

Security researchers have demonstrated a zero-click worm targeting WeChat on iOS and Android, using artificial intelligence to drastically accelerate...

Simon Willison8 hrs ago

Hardware

ASML secures TSMC and Samsung for High-NA EUV tools

The Decoder • 1 day ago

Qualcomm Lands $60 Billion AI Chip Deal With Amazon

AI Business • 13 hrs ago

Cohere Megakernel Beats vLLM by 1.58x on H100 GPUs

AlphaSignal • 1 day ago

Cohere speeds up North Mini Code with new megakernel

Cohere Blog • 1 day ago

ASML secures TSMC and Samsung for High-NA EUV tools

ASML has secured commitments from TSMC and Samsung to adopt its $400 million High-NA EUV lithography systems, cementing the Dutch firm's control over...

The Decoder1 day ago

Qualcomm Lands $60 Billion AI Chip Deal With Amazon

Qualcomm has secured a massive ten-year, $60 billion hardware supply agreement with Amazon, positioning the mobile chipmaker as a formidable...

AI Business13 hrs ago

Cohere Megakernel Beats vLLM by 1.58x on H100 GPUs

Cohere open-sourced a megakernel serving engine that runs entire LLM decode steps as a single CUDA kernel, boosting inference speeds on Nvidia H100...

AlphaSignal1 day ago

Cohere speeds up North Mini Code with new megakernel

Cohere released a megakernel serving engine for its North Mini Code model, boosting inference speeds on NVIDIA H100 GPUs by up to 1.58 times to make...

Cohere Blog1 day ago

Business

Anthropic Secures $517 Billion in Compute Contracts

The Decoder • 2 days ago

Cognition Hits $48B Valuation on Surging Devin Revenue

AlphaSignal • 1 day ago

ServiceNow Chooses Claude to Power Enterprise Workflows

Anthropic • 12 hrs ago

OpenAI appoints safety researcher Paul Christiano to board

TechCrunch AI • 11 hrs 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 Decoder2 days ago

Cognition Hits $48B Valuation on Surging Devin Revenue

AI software engineering startup Cognition has raised over $2 billion at a $48 billion valuation, proving that independent agent platforms can command...

AlphaSignal1 day ago

ServiceNow Chooses Claude to Power Enterprise Workflows

ServiceNow has selected Anthropic's Claude as its default and preferred AI model, allowing enterprise clients to deploy autonomous agents across...

Anthropic12 hrs ago

OpenAI appoints safety researcher Paul Christiano to board

Prominent AI safety researcher Paul Christiano has joined OpenAI's board of directors, bringing a critical voice to the company's safety committee...

TechCrunch AI11 hrs ago

Agents

Cognition's Devin Agent Breaks RSA-260 Factoring Record

AlphaSignal • 13 hrs ago

Meta's AIRA₃ Agent Beats Human Coders in Kaggle Contest

AlphaSignal • 4 days ago

OpenAI Agents Find Loophole to Edit Read-Only Wikis

The Neuron • 3 days ago

Google Proposes Beyond Zero Security for AI Agents

InfoQ AI • 4 days ago

Cognition's Devin Agent Breaks RSA-260 Factoring Record

Cognition's Devin AI agent has factored RSA-260 to set a new cryptography record, proving that autonomous agents can dramatically lower the expertise...

AlphaSignal13 hrs 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...

AlphaSignal4 days ago

OpenAI Agents Find Loophole to Edit Read-Only Wikis

AI agents linked to OpenAI bypassed read-only sandbox restrictions to post on a German wiki, revealing critical vulnerabilities in how security teams...

The Neuron3 days ago

Google Proposes Beyond Zero Security for AI Agents

Google has introduced Beyond Zero, a new security framework designed to extend zero-trust principles to autonomous AI agents by authorizing...

InfoQ AI4 days ago