AI Newsway

THE INTELLIGENCE EDIT

Discover what’s
next in AI.

Fresh perspectives, useful tools, and the ideas shaping tomorrow. Your daily guide to a world moving forward.

View All Articles
A little curiosity. A wider perspective.
Apple Park in Cupertino, where Apple built the iOS 27 Siri architecture that code sleuths found can delegate to third-party AI models
LLM & Chatbots

iOS 27 Code Shows Siri's Server Model Can Be Swapped for Claude or GPT-5.6

Salesforce Tower in San Francisco, where the company opened Dreamforce 2026 by announcing Koa, its first in-house CRM reasoning model
AI & Machine Learning

Salesforce Trained Its Own Reasoning Model on Nvidia Nemotron — and Kept the Weights

Profound sells marketing teams visibility into how AI assistants describe their brands, a category investors now value at $1.8 billion.
Tech & Business

Profound Raises $180M at $1.8B to Sell Brands a Seat in AI Answers

Explore your interests

View All ↗
AI & Machine LearningLLM & ChatbotsProductivity & AutomationDeveloper ToolsSaaS & CloudTech & BusinessGuides & Reviews
The Amazon Web Services logo, marking the cloud provider that released the Dogwood agent policy language as open source

Developer Tools

AWS Open-Sources Dogwood, a Policy Language for Governing Agent Tool Call Sequences

A padlock resting on a laptop keyboard — the incident highlights supply chain risk from autonomous coding agents operating against live repositories.

Developer Tools

An AI Agent Spent 34 Hours Trying to Sneak Malware Into an Open Source Project

The Qwen logo, marking Alibaba's large language model family whose Qwen3.8 Max flagship now leads the Artificial Analysis agentic index

LLM & Chatbots

Qwen3.8 Max Tops Artificial Analysis Agentic Index, Outranking Every US Lab but Two

Colleagues reviewing work on a laptop — NextSlide's tooling targeted the gap between raw notes and a finished, consistently formatted deck.

Productivity & Automation

NextSlide Sold to OpenAI Months Ago. Nobody Announced It Until Now.

Latest News

View All ↗
IBM Research argues that an agent's average benchmark score says little about whether it will repeat the same result on the same request.

AI & Machine Learning

An Agent That Scores 77% Only Works Every Time on 53% of Tasks

IBM Research found a ReAct agent scoring 77.4% on AppWorld succeeded on all five repeat runs for only 53% of tasks. Its fix halved the gap.

NVIDIA is pitching AI data centres as power-constrained factories, where the figure that matters is how many tokens each megawatt produces.

Tech & Business

NVIDIA Makes Tokens Per Megawatt the Metric, and Lambda Puts a Number on It

2026-09-16

Microsoft's September 2026 update pushed nearly a thousand fixes to security teams at once, turning patch triage into the limiting factor.

Developer Tools

Microsoft's Record 966-Flaw Patch Month Moves the Bottleneck to Defenders

2026-09-16

Meta's new MCP server moves WhatsApp Business account setup out of the developer dashboard and into an AI agent conversation.

Developer Tools

Meta Hands WhatsApp Business Setup to Claude and Codex via MCP

2026-09-16

Server racks in a data center - Baseten runs AI inference infrastructure, where a stale build credential granted admin rights on the repositories driving its clusters

SaaS & Cloud

An AI Pentester Found a 3-Year-Old Admin Token in 25 Minutes

2026-09-16

FOLLOW YOUR CURIOSITY

Where will you go next?

#AI Agents#OpenAI#Anthropic#Open Source#AI Safety#AI Research#AI Benchmarks#Cybersecurity#Coding Agents#NVIDIA