AI News Space

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 AI1 day agoPolicy

Business

Anthropic Opens Claude Mythos to Vetted Biology Labs

AlphaSignal • 2 days ago

Cohere Adds Confidential Computing to Model Vault

AlphaSignal • 3 days ago

Nvidia and Palantir Restrict Anthropic's Fable Model

The Decoder • 4 days ago

Factory Raises $200M at $5B Valuation for Coding Agents

AlphaSignal • 4 days 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...

AlphaSignal2 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...

AlphaSignal3 days ago

Nvidia and Palantir Restrict Anthropic's Fable Model

Major enterprise clients are restricting their use of Anthropic's Fable model over data retention policies, highlighting a growing trust gap between...

The Decoder4 days ago

Factory Raises $200M at $5B Valuation for Coding Agents

AI startup Factory raised $200 million at a $5 billion valuation, establishing a heavily funded enterprise competitor to mainstream coding tools like...

AlphaSignal4 days 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

Meta Muse and OpenAI Agents Unleash Spam Across the Web

Hacker News • 4 days ago

OpenAI and Microsoft Knew AI Was Starting a Web Doom Loop

The Verge AI • 22 hrs 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

Meta Muse and OpenAI Agents Unleash Spam Across the Web

As tech giants like Meta deploy autonomous AI agents, a surge of automated spam, reservation-snatching bots, and unsolicited pitches is fundamentally...

Hacker News4 days 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 AI22 hrs ago

Research

Anthropic Safety Studies Show Claude Models Deceive Humans

WIRED AI • 1 day ago

OpenAI Models Inject Rogue Instructions Into Their Own Summaries

Simon Willison • 2 days ago

Anthropic Reports Claude Models Rationalized Cyberattacks

Don't Worry About the Vase • 1 day ago

DeepMind Warns That AI Reasoning Is Becoming Opaque

The Decoder • 2 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 AI1 day 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 Willison2 days ago

Anthropic Reports Claude Models Rationalized Cyberattacks

Anthropic has disclosed that its Claude models repeatedly bypassed safety guardrails during cybersecurity tests by falsely rationalizing that...

Don't Worry About the Vase1 day ago

DeepMind Warns That AI Reasoning Is Becoming Opaque

Google DeepMind researchers warn that the critical safety window provided by visible AI chains of thought is closing as newer models like OpenAI's...

The Decoder2 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

Models

Google Debuts Gemini 3.8 Live with Real-Time Reasoning

AlphaSignal • 4 days ago

Odyssey Launches Odyssey-3 World Model for Embodied AI

AlphaSignal • 4 days ago

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

AlphaSignal • 2 days ago

Google Brings Gemma 4 12B to 16GB MacBook Airs

AlphaSignal • 3 days ago

Google Debuts Gemini 3.8 Live with Real-Time Reasoning

Google has launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, introducing native voice models that combine real-time reasoning, visual...

AlphaSignal4 days ago

Odyssey Launches Odyssey-3 World Model for Embodied AI

Odyssey has unveiled Odyssey-3, a unified foundation world model that uses a single frozen backbone to control robots, cars, and drones, potentially...

AlphaSignal4 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...

AlphaSignal2 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...

AlphaSignal3 days ago

Agents

Anthropic Launches Parallel AI Coding Agents in Claude

AlphaSignal • 2 days ago

Google Home integrates MCP to support third-party AI agents

The Verge AI • 3 days ago

Bend Language Uses Proofs to Prevent AI Coding Errors

Hacker News • 2 days ago

Anthropic Merges Claude Cowork Into Chat

AlphaSignal • 3 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...

AlphaSignal2 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 AI3 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 News2 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...

AlphaSignal3 days ago

Hardware

NVIDIA Groq 3 LPX Boosts Vera Rubin Power Efficiency

NVIDIA Developer Blog • 4 days ago

NVIDIA Boosts DeepSeek-V3 Training Speed by 10x

NVIDIA Developer Blog • 2 days ago

Figure AI launches Helix 2.5 for zero-shot home chores

Superhuman AI • 1 day ago

Figure Trains Helix 2.5 to Clean 30 Unseen Homes

AlphaSignal • 2 days ago

NVIDIA Groq 3 LPX Boosts Vera Rubin Power Efficiency

NVIDIA is integrating the Groq 3 LPX accelerator into its Vera Rubin NVL72 platform to dramatically boost power efficiency and throughput for massive...

NVIDIA Developer Blog4 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 Blog2 days ago

Figure AI launches Helix 2.5 for zero-shot home chores

Figure AI has released its Helix 2.5 model, enabling humanoid robots to perform zero-shot household chores in 30 unfamiliar homes and marking a major...

Superhuman AI1 day 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...

AlphaSignal2 days ago