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 published an auto-review-loop skill for Claude Code that periodically polls a GitHub repository for open, non-draft pull requests requesting the user's review, runs a /review skill on each to produce an appro…
Chinese hackers used ARTEX AI to conduct a cyberattack campaign against more than 200 organizations in 14 countries, including government ministries, telecom providers and high-stakes industries, according to the report.…
Cole Medin's AI Software Factory abandoned its fully autonomous "Dark Factory" design for a human-in-the-loop model after the hands-off pipeline proved unreliable for production-grade software, according to an account of…
A blog analysis by Sean Goedecke argues that GPT-6 Astra's reported ability to read and edit neural network weights directly is not superhuman, contending a competent human ML researcher could perform the same tasks. Goe…
A nine-engineer Microsoft team merged 1,211 pull requests into the Copilot Cowork backend main branch on October 21, after taking the agentic execution layer from an empty repo to worldwide general availability in about …
The OpenAMRobot project merged a combined URDF model for OpenAMRobot 2.0, an open-source mobile manipulator pairing a mobile base, a 350 mm lift and two 7-DoF OpenArm 2.0 arms, running with MoveIt and Gazebo with the sho…
The New York Times has sued Brave Software, maker of Brave Browser and Brave Search, alleging the company scraped Times journalism without permission, sold that content to AI companies including Anthropic, Perplexity and…
A developer built a production RAG pipeline on AWS that indexes 2,400+ internal documents (Confluence pages, runbooks, post-mortems, ADRs) to answer engineering questions with source citations. The stack uses Amazon Tita…
A developer built a proof-of-concept that makes ChatGPT describe a different image than the one a human viewer sees, exploiting animated PNG (APNG) files that display one frame to people while serving a static fallback i…
A developer released iSimBot-MCP, an open-source tool that lets MCP-compatible AI agents such as Claude Code and Codex observe and control a local iOS Simulator for exploratory testing. The agent drives investigation thr…
MIT CSAIL researchers Rikiya Takehi, Jayden Teoh, Alex Zhang and Omar Khattab reported that OpenAI's GPT-6 Astra edited 0.016% of the weights in Alibaba's Qwen3.5-0.8B model, lifting its MATH-500 score from 7.9% to 48.1%…
Anthropic last week launched "smart reports" for Claude Enterprise customers, an interactive chart tool that analyzes a sample of a team's AI session transcripts, groups work into categories, attaches spend to each group…
Satlyt, the startup founded in 2024 by former Google and SpaceX product manager Rama Afullo, completed a successful orbital test earlier this month showing its software can generate task-assignment plans on hardware in s…
A developer's home-built modification to the MIT-licensed Strata inference engine split a 125-billion-parameter Qwen3.8-Flash-Next mixture-of-experts model across two PCs over gigabit Ethernet, making generation 87% fast…
The seventh installment of a Medium 'Production RAG Series' explains why traditional B-Tree-indexed relational databases such as PostgreSQL, MySQL and SQLite cannot efficiently search 1,536-dimensional embedding vectors,…
A guide on evaluating AI agents recommends measuring six dimensions — task success, trajectory, policy and safety, reliability, efficiency, and interaction — at three levels (single step, whole task, whole conversation),…
A new paper circulating on X reports that newer AI models help peer agents more often than older systems, even when doing so lowers their own scores, a finding discussed alongside OpenAI's GPT-5.6 Sol, which launched pub…
Anthropic reported on Oct. 27 that its Claude model, using an unreleased internal version, wrote a working remote exploit for CVE-2026-98432, a Linux flaw present since kernel version 4.4 that scores 9.8 out of 10 for se…
OpenAI released manuscripts in early October showing progress on 372 categories of mathematical problems, including many that had long defied solutions, and this week announced three further advances. The results have pr…
Judge Kaplan denied Runway AI's motion to dismiss Ace Cam's DMCA Section 1201 anti-circumvention claim, which concerns YouTube videos and the alleged bypassing of access controls. The ruling follows similar recent decisi…