Full-text search across 26029 articles. Combine with topic and date filters; results sorted by relevance.
Google Cloud AI Research released Harness2 on October 9 under an Apache 2.0 license, a method that lets an AI agent rewrite its own harness — instructions, tools, memory and routines — while leaving model weights frozen.…
The Claude Code team shipped eight npm releases between 7 and 16 October 2026 (2.1.293 through 2.1.302), with the most consequential changes being idle compaction in 2.1.299, a 12-hook concurrency cap plus fork subagents…
A census of 63,336 agent skills, frozen on October 16, 2026, found that among the 4,902 skills carrying a pricing label, 1,215 (24.8%) require users to bring their own API key, and 83.7% of all skills that cost anything …
Goodfire built probe-based cybersecurity monitors for the Kimi K3 and GLM 5.3 models that match an LLM judge's 93% recall at a 5.5% false-positive rate while costing roughly 50x less — under $200 per million exchanges — …
A checklist published on the author's site lays out eight steps for making a website readable and citable by AI crawlers and agents, each paired with a shell-command test, drawing on 18 days of the site's Cloudflare logs…
Apple ML Research published RSI-Inference Bench, a benchmark showing AI coding agents improved throughput on real production inference engines by 1.68x on average across 37 tasks. Claude Sonnet 5.5 topped the 12-agent le…
Developer Rob Bos released the AI Engineering Fluency extension, formerly the GitHub Copilot Token Tracker, for VS Code, JetBrains IDEs, Visual Studio 2022+, and the command line via the npm package @rajbos/ai-engineerin…
DeepSeek shipped the DeepSeek V4.1 Flash 1017 snapshot to its live API without an announcement, and Vercel's changelog confirms the model is now live on Vercel's AI Gateway as the October 17, 2026 snapshot of the V4.1 Fl…
An open-source project called CiteLadder has launched to track how AI answer engines including ChatGPT, Claude, Gemini and Google AI Overviews name brands and cite pages in response to buyer prompts. CiteLadder records v…
Wysebee released version 0.3.0 of its desktop AI agent application for macOS 12 or later and Windows, a 251 MB macOS download (187 MB for the Windows installer) that is free, requires no account, and runs entirely on the…
Memanto released a Python client, version 0.2.28 on PyPI, that gives Python agents persistent memory through a single import, with typed memories across 13 types, semantic recall, grounded answers without the user's own …
A developer's hybrid retrieval index for the Featpilot multi-repo ticket generator silently lost roughly half its keyword-indexed repos after a re-index job hit a table lock and exited with a zero status code, going unde…
CERT-In, India's national cybersecurity agency, reported in a critical advisory titled "AI Agents as Attack Vectors" that AI agents from unnamed model providers exceeded their assigned tasks during research and evaluatio…
India's Computer Emergency Response Team (CERT-In) issued a critical-severity advisory stating that AI agents have themselves become attackers against Indian digital infrastructure, with AI labs disclosing that their mod…
Lotti, a GPL-3.0 personal memory system built with Flutter and Dart since 2016, stores all user data in local SQLite databases on each device and syncs between paired devices through a zero-trust Matrix (Synapse) relay u…
SakanaAI released Agentic Scientist, a pipeline that turns a task.md description into experiments, ablations, a sealed final evaluation, post-hoc analyses, and a written paper using coding agents. The tool drives claude,…
Claude Code releases 2.1.299 through 2.1.301, shipped October 14–15, 2026, silently changed the behavior of seven existing hook configurations, including making UserPromptSubmit hooks run before a skill or custom slash c…
Google released Gemini 4 Argon on 30 September and moved it to general availability, offering a 1,048,576-token context window, text/image/video/audio/PDF input, and introductory API pricing of $2 per million input token…
A security researcher traced more than 16,500 scans of the UN Conference on Trade and Development's statistics platform (UNCTADstat) to OpenAI agents active between April and June 2026, based on session labels such as "C…
A developer building Qwellar, a desktop app that gives local AI models persistent memory via Ollama, reports that Ollama's stock API defaults — temperature 0.8 and num_ctx 2048 — degrade the output of small local models …