The New Stack
What it covers
Latest stories it covered
Alibaba launches Qwen4-Max, a closed flagship AI model with lower API costs and 1M context
Alibaba has unveiled Qwen4-Max, its newest flagship large language model in the Qwen4 family, and makes it generally available via its QwenCloud API. Multiple outlets say the compa...
Google says attackers used AI agents to convert security patches into working exploits
Google says it has observed adversaries using AI in a more autonomous way than previously seen: turning security patches into exploits without human involvement. In a new report fr...
AWS releases open-source AWS Emulator for local testing of 200+ cloud services
Amazon Web Services has released AWS Emulator, an open-source tool that provides a local HTTPS endpoint for testing against more than 200 AWS services without using an AWS account....
Anthropic releases Claude Fable 5.2, comparing it with GPT-6 Astra and open models
Anthropic releases Claude Fable 5.2 on 9 November 2026, alongside Claude Mythos 5.2, described as the same model variant with fewer safeguards for vetted users. Coverage across out...
Google launches Gemini 4 Hydrogen AI model for faster, agentic tasks and improved Search
Google has released Gemini 4 Hydrogen, a new AI model presented as faster than its predecessor and positioned for coding, research, and “agentic” workflows. Multiple outlets say it...
Anthropic launches Claude Fable 5.2 and Mythos 5.2 with different safeguards and lower prices
Anthropic has released two new versions of its Claude models: Claude Fable 5.2 and Claude Mythos 5.2. Multiple outlets describe them as based on the same underlying model but deliv...
Google launches Gemini Decisions API with Gemini Decision 1 for fast multimodal decision-making
Google has launched the Gemini Decisions API, powered by a new model called Gemini Decision 1, aimed at handling “closed-set” decisions in applications instead of generating prose....
OpenAI unveils GPT-6 Cyber, citing Firefox flaw discoveries and Linux exploit development
OpenAI unveils GPT-6 Cyber, a cybersecurity-focused AI model it says is designed to help authorized researchers find software vulnerabilities, develop exploits, and support fixes....
Perplexity launches always-on AI “Teammates” for work tasks
Perplexity launches “Teammates,” an always-on set of work-focused AI agents designed to take on ongoing tasks for users. Outlets describe the agents as being given fixed roles, abl...
Goldman Sachs restricts most staff’s access to Claude Code and GitHub
Goldman Sachs restricts most traders and bankers from using AI coding tools including Anthropic’s Claude Code and Microsoft’s GitHub, limiting access mainly to engineers and some q...
Dropbox to acquire Tigris Data for $350M and launch DBX Agent Cloud
Dropbox is acquiring cloud storage startup Tigris Data for $350 million, and announcing the deal alongside new AI-oriented product updates. Multiple outlets say the acquisition is...
Anthropic updates Claude Code to prevent AI recreations of existing video games
Anthropic updates its Claude Code AI coding tool to block users from recreating specific existing video games, even if renamed, and tells the model to focus on original work. Multi...
OpenAI merges ChatGPT’s Chat and Work tabs and shares ChatGPT-Codex usage limits
OpenAI is changing how ChatGPT is used by planning to merge the Chat and Work tabs into a single experience. Multiple outlets report that the update aims to let users move between...
Canva launches Brush, a vision-language AI model for editable design refinements
Canva has unveiled Canva Brush, a vision-language model aimed at helping designers edit and refine existing work inside the Canva platform. Unlike image generators that create a co...
OpenAI launches GPT-6.1 Terra, a cheaper model for everyday computer tasks
OpenAI has launched GPT-6.1 Terra, a mid-tier model positioned for routine professional “computer use” work such as writing, visual understanding, and operating desktop application...
Google DeepMind previews Genie 4, an AI world model with real-time interactive audio and visuals
Google DeepMind is rolling out a research preview of Genie 4, an upgraded “world model” that turns text prompts into interactive, photorealistic environments. Multiple outlets desc...
PHP2Shell RCE flaws exploited against Palo Alto Networks management interfaces and other targets
Multiple security outlets report that attackers are exploiting PHP2Shell—two newly patched critical PHP remote code execution (RCE) flaws, CVE-2026-109618 and CVE-2026-109619—to co...
Nvidia releases open Nemotron 3.5 Transcribe for multilingual audio transcription and translation
Nvidia releases Nemotron 3.5 Transcribe, an open-weights speech model designed to convert audio into text and support multilingual translation. Multiple outlets say it can handle l...
Mistral releases open weights for Large 4 model on Hugging Face
Mistral AI SAS has released the open weights of its frontier language model, Mistral Large 4 (ML4), also nicknamed “Le Chonk.” The model, initially previewed about three weeks earl...
PHP releases emergency fixes for critical unauthenticated RCE flaws dubbed PHP2Shell
PHP releases emergency patches for two critical vulnerabilities in PHP core that can allow unauthenticated remote code execution (RCE). The issues, dubbed “PHP2Shell,” are assigned...