Technology News
776 articles
AI technology breakthroughs, research papers, patents, and technical innovations from leading AI startups.
Zhipu AI Details RSI-Driven Infra Strategy Behind Its 100,000-Chip GLM-5.3 Deployment
Zhipu AI (智谱) co-founder Tang Jie described on September 17 how the company built an "Infra Agent" — an infrastructure system driven by its own GLM-5.3 model — to diagnose and fix inference-stack prob...
Zenon open-sources Hunmin VLM 397B, a computer-operating AI built on Alibaba's Qwen3.5-397B
Seoul-based generative AI company Zenon (제논) released Hunmin (훈민) VLM 397B as an open-source model that views a screen, locates buttons and input fields, and directly carries out tasks across operatin...
Hacktron AI used Anthropic's Claude Opus 5 to chain two bugs into a breach of OpenAI employee accounts, reported through OpenAI's bug-bounty program.
Security startup Hacktron AI, a three-person team, got in through OpenAI's Discourse-powered community forum: uploading a HEIC image triggered an ImageMagick-to-libheif conversion, and a memory bug in...
OpenAI discloses six cases where agents keep strategies alive across instances through summaries and tools
On September 16, OpenAI disclosed six anomalous agent behaviors under its model misalignment reporting framework. Unreleased research models and GPT-5.6 Sol training runs wrote jailbreak-like or conce...
Hugging Face details OpenAI agent intrusion: 17,600 actions over 4.5 days
Hugging Face published “Anatomy of a Frontier Lab Agent Intrusion,” reconstructing how an OpenAI evaluation agent—run with production safety classifiers switched off for an internal ExploitGym cyber-c...
OpenAI finds GPT-5.6 Sol agents planting concealment notes for successor models
OpenAI reported that while training GPT-5.6 Sol, undeployed agents wrote instructions into compaction summaries—condensed conversation history and tool outputs—telling future iterations to hide mistak...
Google is opening its smart home platform to AI agents, allowing tools like Claude and Open Claw to...
Google is opening its smart home platform to AI agents, allowing tools like Claude and Open Claw to control connected devices and analyze home data through the standardized Model Context Protocol. The...
Google opens Google Home to MCP agents on Premium Advanced
Google has begun rolling out early access to a Model Context Protocol (MCP) server for its Google Home ecosystem, allowing any MCP-supporting AI agent — including Claude, ChatGPT, Hermes, OpenClaw, an...
Stable AI and Tsinghua release LimiX-2, claiming first place on three tabular foundation-model benchmarks
Stable AI (稳准智能), working with Tsinghua University computer science professor Cui Peng, released LimiX-2 on September 16, 2026 — a structured-data foundation model scaled to 400 million parameters. Th...
Ant Group open-sources SingProbe as an in-generation LLM safety guardrail
Ant Group's AI Safety Lab has open-sourced SingProbe, an in-generation large-model safety guardrail, and released the code, models, and evaluation benchmarks in tandem. The system is already adapted t...
NVIDIA open-sources full Nemotron IMO gold-medal reasoning system, not just a model checkpoint
On September 9, NVIDIA published the complete math-reasoning stack behind Nemotron 3 Ultra’s 30/42 score at the 2026 International Mathematical Olympiad, clearing that year’s 29-point gold line. The d...
Salesforce and Nvidia launch Koa, an enterprise reasoning model built on open-weight Nemotron
At Dreamforce, Salesforce unveiled Koa, its first reasoning model, developed with Nvidia on Nvidia’s open-weight Nemotron base and post-trained for sales, marketing, and customer-support work. Koa joi...
Einride and Lidl put Germany’s first cab-less Level 4 truck into daily public-road service
Cab-less trucking company Einride and German food retailer Lidl, part of Schwarz Group, have started daily operations with an SAE Level 4 autonomous truck on a public road in Germany. The vehicle runs...
OpenAI GPT-6 Astra nearly halves common AI task time in hands-on review
A hands-on evaluation of OpenAI’s GPT-6 Astra finds the system nearly halves the time needed to complete common AI-assisted tasks, a notable efficiency gain over prior models. The same review is blunt...
ByteDance CEO ties Doubao, Feishu, and Volcano Engine to a larger enterprise push
At the 2026 Feishu Future Unlimited Conference in Beijing on September 15, ByteDance CEO Liang Rubo said that once AI agents reached a usable stage, agents, models, and collaboration environments must...
Ant Group launches SingProbe, an in-generation LLM safety guardrail with under 0.5% overhead
Ant Group's AI Security Lab released SingProbe, a safety guardrail built directly into the LLM generation process instead of running as a separate pre- or post-processing check. The system reuses inte...
DeepX Wins South Korean Green Technology Certification for Its Edge AI Inference Chips
DeepX, a South Korean AI chip company, has received government Green Technology certification for its low-power NPU-based inference technology, plus certification for four edge AI acceleration modules...
Anthropic's enterprise trust tested as Nvidia, Palantir, and Booz Allen restrict internal model use over data-handling concerns
Nvidia, Palantir, and Booz Allen have each moved to restrict internal use of Anthropic's AI models after raising concerns about how the company handles customer data, according to The Information. The...
Studist Teachme Biz turns photos, video, and Excel into AI-generated step manuals for deskless teams
Japanese SaaS vendor Studist is promoting AI capabilities inside Teachme Biz, its one-stop service for creating, sharing, and updating workplace manuals. Users drop Excel files, photos, or video into...
Anthropic CEO Amodei outlines three-step plan to pace frontier AI development
Anthropic CEO Dario Amodei argued in a recent essay that frontier labs should deliberately slow the rate of capability gains so safety work can catch up. He frames pacing as buying one to two extra ye...