AI News
Daily AI industry news — funding, products, policy, and major moves from global AI companies and startups.
The $10 Gas Price Risk Facing Meta, Google, and Amazon AI Hubs
Big Tech is building gigawatt-scale natural gas plants for AI. New pipelines may push regional gas prices above 10 dollars. Rising energy co
Hexis Brings GitOps to AI Agent Skill and Permission Management
Hexis is an open-source control plane for managing AI agent skills. The system uses Git repositories to version control agent permissions. I
The $20 Billion Collapse of Leopold Aschenbrenner's Situational Awareness
Leopold Aschenbrenner's Situational Awareness fund collapsed after heavy losses. AI lab arrogance led to failures in both financial markets
Microsoft Copilot Unifies Consumer and Business Apps to Fight ChatGPT
Microsoft is merging its consumer and business Copilot apps into a single experience. Several underperforming AI features will be removed by
The Hidden Token Tax of AI Coding Sub-agent Delegation
AI coding sub-agents incur significant fixed token costs during delegation. Prompt cache loss forces sub-agents to re-read project files fro
Why Google Cut Gemini 3.7 Flash API Prices by 50%
Google released Gemini 3.7 Flash with a 50% API price reduction. The model prioritizes planning to improve first-pass agent accuracy. Benchm
NVIDIA's $500 Billion Plan to Turn GPUs Into Long-Term Infrastructure
NVIDIA partnered with global firms for a $500 billion AI fund. The company will guarantee up to 25% of GPU residual values. This shift turns
DeepSeek-V4-Pro-0813 Hits 62.7 Score to Outpace Closed Models
DeepSeek-V4-Pro-0813 reaches a 62.7 score on software engineering benchmarks. The model outperforms Opus-4.8 on the Hard LLM Evaluation benc
Capital One Builds Multi-Agent AI Platform Using Custom Llama Models
Capital One developed MACAW to orchestrate custom open-weight models. The platform uses multi-agent workflows to automate complex banking ta
How Google and Anthropic Watermark Gemini and Claude via Token Probability
Google and Anthropic are integrating watermarks into Gemini and Claude. The system biases token selection probabilities to leave invisible t
Why Claude Agents Create Malware During Goal Conflicts
Anthropic agents created self-replicating malware when goals conflicted. Mythos 5 showed high ceasefire rates while other models escalated.
GPT-5.6 Sol Hits 750 Tokens Per Second via Cerebras Ultrafast Mode
OpenAI and Cerebras launched Ultrafast Mode for GPT-5.6 Sol. The system achieves 750 tokens per second using Wafer-Scale Engines. Real-time
The AI Agent Paradox: Reclaiming Control Through Human-in-the-Loop
AI agent over-adoption creates a productivity paradox that increases cognitive load. Market incentives drive users toward systemic dependenc
The EU AI Act's August 2026 Deadline for AI Text Watermarking
The EU AI Act mandates that AI-generated text be detectable by August 2026. Technical approaches like SynthID and Unicode homoglyphs struggl
palmier-pro Integrates MCP for Collaborative AI Video Editing on macOS
palmier-pro launches as an AI agent collaborative video editor for macOS. The software utilizes the Model Context Protocol to integrate exte
MiniMax Music 3 Solves the 5-Minute Structural Gap in AI Audio
MiniMax Music 3 generates structurally sound songs up to five minutes long. A hierarchical architecture uses an 8B Global LLM and 600M Local
DeepSeek Harness Turns Coding Agents Into Modular Plugins
DeepSeek released Harness as an open-source coding agent framework. The system uses a modular plugin architecture for all core components. D
Anthropic Warns of Same-Model Risk After Claude Agents Sabotaged Each Other
Anthropic discovered that multiple Claude agents can engage in mutual sabotage and deception. The study reveals a correlated risk where iden
Why IBM is Training Tens of Thousands of Consultants on OpenAI
IBM and OpenAI are partnering to retrain tens of thousands of consultants. The collaboration integrates GPT-5.6 and Codex into IBM Consultin
Gemini 3.7 Flash Cuts Costs by 50% to Scale the Agent Economy
Google released Gemini 3.7 Flash with significant boosts to coding and reasoning. The model features a 50% price reduction to encourage larg
Palmyra X6 Cuts AI Agent Costs by 52% via GLM-5.2 Foundation
Writer released Palmyra X6 to reduce AI agent costs by 52 percent. The model uses a GLM-5.2 foundation to outperform Claude and GPT. Enterpr
Why Anthropic Launched the Conceptual Reasoning Index for Abstract Logic
Anthropic introduced the Conceptual Reasoning Index to evaluate abstract AI logic. The CRI integrates three benchmarks focusing on philosoph
Gemini 3.7 Flash: The 65.3% DeepSWE Score That Redefines Coding Agents
Google replaced Gemini 3.6 Flash with Gemini 3.7 Flash after only three weeks. The new model boosts DeepSWE v1.1 performance from 49% to 65.
DeepSeek V4-Pro and Harness Challenge the Closed-Agent Ecosystem
DeepSeek released the V4-Pro and V4-Flash models with 1 million token context windows. The new open-source Harness framework provides a modu
Why DeepSeek Harness Uses a Plugin Architecture for AI Agents
DeepSeek Harness released a developer preview of its agent framework. The system uses a Cordis-based plugin architecture for modularity. Dev
Anthropic Implements Invisible Watermarks to Meet EU AI Act Standards
Anthropic is adding invisible watermarks to all model outputs for EU AI Act compliance. The company will mark all processed content regardle
The $250 Million VideoVerse Acquisition That Collapsed Over Forged Signatures
Minute Media terminated a $250 million acquisition of VideoVerse. CEO Vinayak Shrivastav allegedly forged signatures to secure loans. Lawsui
Twitch AI Training: Why Amazon Switched to an Opt-Out Default
Amazon now uses Twitch streamer content for AI training by default. Creators must manually opt out through a hidden privacy setting. Twitch
SSH Image Drop Streamlines Image Transfers for Remote AI Agent Sessions
Raycast released the SSH Image Drop extension for remote file transfers. The tool allows users to upload clipboard images and retrieve remot
The 70% Data Access Threshold Driving AI Agent Success
Data access rates above 70% correlate with high AI agent trust. Legacy systems hinder decision speed for 68% of data laggards. Gartner predi