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's PageBreak AI security agent has uncovered more than 500 XSS vulnerabilities across the company's web applications since entering full operation in January, according to a paper Google published. PageBreak uses G…
Anthropic reported that a Claude model, given enough attempts in a deliberately one-sided adversarial game, succeeded in writing fine-tuning data that made a small research model misaligned while evading a training-data …
A developer built Ofrenda Digital, an offline-first digital altar for their late grandmother, Nenek Lia, as a Hacktoberfest Week 4 "Ghost in the Machine" submission. The single-file vanilla HTML/CSS/JS app stores memorie…
A developer built Abuela's Recipe Cards, an offline-first single-file web page that presents a family strawberry pumpkin brownie cake recipe as a candlelit Día de los Muertos ofrenda. A locally run open-weight model flag…
A developer building a retrieval-augmented generation (RAG) assistant for document processing replaced a split SQLite-plus-Chroma store with a single Postgres database using a pgvector column and a generated tsvector col…
A paper posted to arXiv on 2 June 2026 and revised 10 July 2026 proposes a "Sleep" paradigm that lets large language models continually learn by distilling short-term in-context memories into long-term parameters. The tw…
Developer Tapaskar released archwhere, an open-source tool that puts every AI coding agent's pull request on a declared architecture map and requires agents to keep that map current as a condition of merging. archwhere s…
An analysis of 63,300 Hacker News comments and posts from 2008 to 2026, filtered by an AI classifier to 12,441 software requests, identified the 100 tools users most often ask for, backed by 1,433 re-read asks and $6.00 …
Google's Gemini 4 Carbon (high) topped the updated AA-Omniscience v1.1 benchmark from Artificial Analysis with a Net Score of 54, 6 points ahead of second-placed Claude Opus 5.5 (48, max with fallback) and 8 points ahead…
Mistral AI released the open weights of Mistral Large 4, nicknamed Le Chonk, on October 29, 2026, and the model scored 48 on the Artificial Analysis Coding Agent Index, matching OpenAI's GPT-6.1 Terra and Codex, accordin…
A developer built a per-message RAG "memory recall" hook for Claude Code that embeds past notes and conversations with BAAI's bge-m3 model and injects the closest matches into context before each reply, capped at 2,048 b…
A university computer science instructor in Bengaluru built Séance, an open-source terminal tool that uses a local Gemma 3 4B model served through Ollama to role-play dead programming languages — COBOL, FORTRAN, Pascal, …
A developer built Shezhire, an open-source pipeline that turns a recorded Kazakh family story into a genealogy (шежіре) using Gemma 4 via Ollama, with a verification layer that requires every proposed family link to be g…
A new arXiv paper (2610.37830v1) finds that AI agents writing smart contracts lose requested behavior between plain-English intent and deployed code, and that asking clients clarifying questions before generation outperf…
A new arXiv paper (2610.37739v1) introduces OpenFrontiers, a cross-disciplinary corpus of 12,151 open research problems paired with a retrospective benchmark of 200 mathematics problems, 110 resolved in 2026 and 90 still…
A new arXiv paper (2610.37849v1) proposes Partner-Outsider Positive Listening (PoPLi), a behavioural test that detects covert collusion between AI agents by comparing how much a partner agent's actions change versus an u…
A new arXiv paper (2610.37851v1) introduces LibStructBench, a document-grounded benchmark for testing whether AI agents can reconstruct the exact sequencing-library structures and molecular workflows specified by sequenc…
A new arXiv paper (2610.37859v1) introduces Seq@K, an evaluation protocol that measures whether a language model can solve a problem within K sequential attempts when given feedback after each prior failure. The authors …
A new arXiv paper, Verification Learning Theory (arXiv:2610.37875v1), proposes a formal process model linking perceived AI assistant reliability to evidence checking and linking the history of those checks to retained ru…
A study posted to arXiv (2610.37896v1) finds that socio-demographic information yields small and inconsistent gains for individual-level offensiveness predictions from prompted and trained LLMs, but clearer gains at the …