Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
A developer built a Magic: The Gathering benchmark that pits large language models against each other through Forge, an open-source Java MTG rules engine, after finding that early agent-vs-agent tournaments exposed harne…
A developer outlined an approach to instrumenting agentic AI systems with OpenTelemetry, using traces, metrics and logs to follow multi-step agent workflows such as LLM calls, tool invocations and subagent delegation. Th…
Anthropic added a two-sentence rule to its Claude Code programming tool instructing the AI not to build "a recreation of a specific existing game, even if it is renamed or called original," a change Anthropic software en…
OpenAI released 722 mathematical manuscripts grouped into 372 families, produced by an internal AI model, according to a blog post by mathematician Gil Kalai. Kalai called the release "an amazing milestone for mathematic…
NVIDIA's AI Cluster Runtime (AICR), part of the NVIDIA DSX AI Factory Platform, will let Kubernetes environments run NVIDIA Vera Rubin NVL72 compute trays on Day 1 by capturing tuned configurations in reusable recipes, g…
Diffio.ai, which builds a speech restoration tool, reported that in October 2026 it ran 59 real recordings through six speech cleanup tools and found five of seven tool configurations raised word error rate for a small s…
A team at Cal Hacks 13.0 built Silent Voice, an open-source system that converts silent mouthing into real-time speech in the user's cloned voice, using roughly $20 EMG sensors, a webcam, Claude and Cartesia Sonic 3.6; t…
OpenAI announced LifeVal on November 2nd, an internal benchmark that grades AI models on everyday personal tasks such as finding childcare, creating a budget, ordering groceries and accessing government services, and sai…
GitHub is publishing status updates for an incident titled "Elevated errors on GPT-5.4 Nano," offering email and SMS notifications when the incident is updated, created or resolved. The notice does not state the error ra…
Contextual AI researchers Sheshansh Agrawal, Thien Hang Nguyen and Douwe Kiela released BlitzRank, a tournament-graph framework for zero-shot LLM reranking that achieves Pareto dominance over existing methods across 14 b…
Google released Gemini 4 Argon globally in October 2026 across Google AI Studio, the Gemini API, Google Cloud, Search and the Gemini app, then announced Gemini 4 Carbon two weeks later as its strongest model yet for long…
Aylon launched ArchDocs, a tool that adds shared context and verification for engineers and AI coding agents, after starting the project in late 2024 as an architecture diagramming tool. ArchDocs' new Design Docs workflo…
Tel Aviv startup 42AI emerged from stealth on November 2 with a "behavioral world model" it says is already running inside 10 brokerages and reading close to 30 million people, backed by $62.5 million built up over three…
GitHub migrated its Copilot runtime from TypeScript to Rust using Copilot agents, landing the work through 128 pull requests at a cost of approximately $120,000 in tokens, according to a GitHub blog post by Stephen Toub.…
A from-scratch Rust browser engine called Tab passed 1,774,103 of 1,912,797 upstream web tests (about 95%) in build 0.1.2681, measured October 26, 2026, according to results published by the project. The engine, which us…
The Subvocal Company, founded by Aarush and Arnav out of Georgia Tech and backed by Y Combinator's F26 batch, is building smart glasses that decode silently mouthed words into text at the speed of speech using radio-freq…
A five-workload benchmark of newly tested models found GPT-6 Sol 6 cheaper and faster than GPT-5.6 Sol with similar quality, scoring 0.944 mean quality at $0.036 per task and 12.2 seconds latency versus 0.939, $0.081 and…
A new blog post proposes a "social harness" for agentic societies, a layered stack of capabilities modeled on the Internet's networking stack, to manage interactions between AI agents across trust boundaries. The post ar…
A project at elections.lforla.org published a public methodology for scoring AI election forecasts on reasoning quality rather than outcomes, rating each pre-vote answer 0 to 100 on specificity, grounding and calibration…
CentOS and Rocky Linux creator Gregory Kurtzer built an experimental operating system kernel called Ginux in 11 days using OpenAI Codex, powered by GPT-5.6 Sol, under a process he calls Human-Directed Agentic Engineering…