Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
Google announced it plans to voluntarily adopt an IETF AIPREF draft proposal that would let publishers express AI training preferences across multiple AI providers through a single shared "policy token" in robots.txt, wi…
Devansh Gupta built The Last Lantern, an offline-first AI storytelling web app that runs entirely on a locally hosted open-weight model. The static HTML/CSS/JavaScript frontend calls Ollama's local API with the Qwen 2.5 …
Product Hunt's daily launch volume on the second Wednesday of May jumped from 67 launches in 2023 to 139 in 2024, 258 in 2025 and 783 in 2026 — roughly 17x growth over three years — according to a manual count of the sit…
PostgreSQL 19 will likely ship in November rather than the usual mid-September to mid-October window, and the project's committers attribute the delay to LLM-driven bug discovery. Between feature freeze on April 8 and Se…
OpenAI unveiled Private Intelligence at its DevDay event on September 29, a package pairing available zero-retention safety processing with a private inference preview aimed at letting businesses run its most capable mod…
Researchers posted WxCrop to arXiv on 29 October 2026, describing it as the first agent that judges agricultural hazards from physically interpretable weather indices examined by causal analysis and the first to combine …
Y Combinator and Olam Labs ran three multi-agent experiments in which 50 GPT-6.1 Sol agents spent 3 hours in one Minecraft survival world, sending about 5,900 messages and dying 34 times, 22 of those on the first night, …
A team led by El Mustapha Mansouri posted an arXiv paper on 29 October 2026 describing an autonomous research laboratory in which large-language-model agents review one another's work and can halt or redirect investigati…
Anthropic reported on October 27 a remotely exploitable use-after-free vulnerability, tracked as CVE-2026-98432, in Linux multipath TCP listeners, where a retransmitted SYN packet races connection setup in another thread…
An AI engineer describes building an explainable causal reinforcement learning framework for satellite anomaly response, after a PPO-trained agent burned through a simulated satellite's battery margin in under 40 minutes…
A developer built RepoSéance, an open-source local code-investigation tool that uses the Gemma model (gemma4:e2b) via Ollama to explain forgotten or abandoned source code. The full-stack JavaScript app, built on Node.js …
Microsoft Fabric and Neo4j Graph Intelligence combine to give fraud-fighting Copilots relationship-level context by converting flat Delta tables in OneLake into connected property graphs, according to Microsoft. The comp…
Researchers at Google-owned Wiz and an AI firm called Calif used models from OpenAI and Google to discover critical PHP vulnerabilities, dubbed PHP2Shell, that let attackers take over targeted websites running on PHP, wh…
A developer built PHANTOM, an open-source, offline-first AI presence for the terminal that runs local models through Ollama (defaulting to qwen3:8b) with SQLite-backed persistent memory and a deterministic mood system ti…
A developer built Dead Code Ouija, a local-first, client-side tool that uses custom AST-pattern heuristic matching to detect legacy JavaScript constructs such as var declarations, outdated DOM selectors and callback pyra…
Researchers introduced ABF-Bench, a verifiable benchmark for agentic blockchain forensics that tests AI agents on adversarial, open-ended tracing tasks in which launderers split, swap, and bridge funds and cut trails at …
Researchers report that the staged pipeline used to train foundation models — pretraining, midtraining, supervised fine-tuning (SFT) and reinforcement learning (RL) — is not algorithmically fundamental, and that suitable…
Daniel Catt built a markdown-to-handwriting tool, documented in his newsletter at newsletter.revdancatt.com, that drives a Bantam Tools NextDraw 8511 pen plotter through Kitty, his AI assistant. Kitty writes out Catt's m…
An audit by AI agents working under MV2 at Munim, Inc. analyzed all 48 published MMLU-Pro prediction files and found that, across 45 models matching the current test set, every model scores higher on the 656 STEM-website…
Artificial Analysis's Intelligence Index v5 revision raised the weighting of private tests and held-out answers to 60% from 45%, reshuffling the leaderboard so that Anthropic and OpenAI hold the top five places, with Opu…