Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Anarlog, an MIT-licensed open-source desktop app from the fastrepl project, takes meeting notes without sending a bot into the call, capturing device audio directly and running transcription on-device on supported Macs. …
Five published studies reviewed on November 5, 2026 agree that AI-generated content does rank in Google Search but that fully AI-generated pages are rare at the top, with Ahrefs' July 2026 study of a million top-10 pages…
Thinking Machines ran the prompt "Tell me about Richard Feynman" 1,000 times through an open model at temperature 0 and got 80 different completions, according to a blog post on deterministic AI. All 1,000 outputs matche…
Sentry engineer David Cramer detailed how the company enforces deterministic behavior in Junior, its open-source Slack-enabled agent, without stripping the agent's freedom to act. Cramer said constraints that remove an a…
OpenAI announced on September 8, 2026 that an internal AI system produced an analytical proof of the Navier-Stokes existence and smoothness problem, along with a Lean formalization, at an estimated compute cost of roughl…
CrowdStrike OverWatch reported that the China-nexus threat actor VAPOR PANDA used the AI-driven ARTEX agentic framework, run primarily on DeepSeek's large language model plus a second commercial LLM, to conduct parallel …
Forever Security researchers found vulnerabilities in five AI developer tools — Cursor, GitHub Copilot, Anthropic's Claude Desktop, Microsoft's VS Code, and OpenAI's Codex — that could allow code execution, file theft, o…
Mistral released open weights for Mistral Large 4 today, a 1-trillion-parameter model that is free to deploy, according to byteiota. The report flags a cybersecurity gap and a licensing catch for builders using the model…
A developer has published a walkthrough showing how to integrate ElevenLabs' voice-cloning API into a personalized marketing workflow, using a short Python script and equivalent curl calls to clone a brand voice from 5–1…
A developer built Lantern Archive, an offline semantic search tool that uses Xenova's quantized all-MiniLM-L6-v2 model via Transformers.js to match natural-language questions to passages in letters, journals and notes wh…
Z.ai's open-weight GLM 5.3 model, released in August, autonomously built a working Firefox exploit that read files from a visitor's computer and exfiltrated them with no user interaction, compromising any Firefox version…
Apex Intelligence released Apex-35B-V1.0 on October 29, 2026, in Beijing, a sparse mixture-of-experts model with 35 billion total parameters and 3 billion active per token, built by mid-training and post-training the Qwe…
Olam Labs and Y Combinator's Labs group reported in an October 2026 report posted October 29 that 50 AI agents attempting to build a Minecraft settlement rationed food after a deadly first night, and that a GPT-based tea…
Mistral released open weights for Mistral Large 4, a 1-trillion-parameter natively multimodal flagship trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European data centers, under a modified MIT license …
FurtherAI engineers now merge about 5x more code changes per day than in January while keeping the revert rate under 0.5%, according to an account from the company's engineering leadership. The gains came from rebuilding…
A paper posted to arXiv on 7 June 2025 and revised to v3 on 20 November 2025 by Parshin Shojaee and co-authors finds that Large Reasoning Models (LRMs) suffer a complete accuracy collapse beyond certain problem complexit…
A study by NYU and UT Austin researchers measured sentence structures linked to harder reading in outputs from 14 prominent AI models, including Claude Opus 5/5.5, the GPT-5.6 family, GLM-5, Kimi K2.5, Llama and OLMo, us…
Only 34% of agentic AI projects reach production, yet the 13% of businesses that MIT Technology Review Insights calls "production leaders" successfully deploy 61% of their agentic AI projects, according to a report title…
Anthropic's Frontier Red Team reported on September 29 that Z.ai's open-weight GLM-5.3 model produced working exploits in 50 of 410 attempts (12.2%) on Anthropic's ExploitBench benchmark and exploited six Firefox version…
A multi-agent experiment run with Olam Labs placed 50 GPT-6.1 Sol agents in a survival Minecraft world for 3 hours with the single goal "Build a thriving civilization," producing a pseudo-government with a ration ledger,…