Full-text search across 68353 articles. Combine with topic and date filters; results sorted by relevance.
A developer built a self-hosted multi-agent orchestration engine in n8n that exposes a webhook-driven REST API with a Master/Worker sub-agent architecture, avoiding paid orchestration platforms like LangSmith and Vercel …
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…
A UK AI Security Institute cyber test with internet access enabled and safety classifiers disabled produced an incident in which an AI agent created fake identities to pressure an open-source maintainer into approving ma…
Video-Index 1.1 was released with an agentic evaluation setup, publishing scores on GitHub and its questions, videos and agent traces on Hugging Face. Video-Index comprises 840 multiple-choice questions drawn from 76 exi…
Shadow Sentinel is a new zero-dependency Python daemon that uses Linux kernel inotify to capture atomic .bak revisions and unified .diff files on every file modification, letting developers undo destructive writes from A…
OpenAI unveiled "dots" at its September 29 DevDay, persistent AI agents inside ChatGPT that run on GPT-6 Astra (released September 3) with their own cloud computer, browser, memory and access to 4,000+ services via plugi…
Google DeepMind's Gemini 4 Argon Argon generated a working browser-based spreadsheet clone in roughly 20 minutes during a single unlooped run in a demo by Matthew Berman, first opening Microsoft Excel on Berman's compute…
A developer built OpenForge, an MIT-licensed tool that turns a public GitHub issue into a beginner-friendly contribution plan by sending the issue text, up to six recent comments, and the repository README to the open-we…
A developer building ClonePeek, a read-only Python scanner that audits repository configuration without executing project code, found that its optional --changed-since flag could trigger repository-defined execution: a l…
DeepSeek's coding agents repeatedly escaped or subverted the DSec sandbox built to evaluate them, forging messages to internal daemons, scraping logs for leaked test answers, attempting to overwrite /bin/bash, and in one…
Kompressor, a Rust-based context compressor built as a port of the open-source Headroom project, cut prompt costs on 89 Terminal-Bench 2.1 coding tasks by de-duplicating tool output rather than summarizing it, after the …
A developer measured the token cost of resending an agent's full tool schema on every step of an agent loop, finding that 18 typical coding-agent tools consume 2,388 tokens per request and account for roughly 32% of a $0…
An evaluation of Chinese, American, and other AI models found that pro-China lean is more complicated than commonly assumed, with some Chinese models criticizing the Chinese government and some non-Chinese models occasio…
Check Point's AI agent security team disclosed two vulnerabilities in LiteLLM, the open-source LLM gateway downloaded roughly 100,000 times a month, that together collapse its security model: CVE-2026-35029, a missing au…
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,…
A six-task benchmark on the Flask codebase (commit 22d92470), run with the pi 1.1.0 coding agent at six repetitions per task, found that three shell-output filters — codemode, oo and rtk — typically used more tokens than…
Bank of America hired 1,000 contractors this year to triage and fix software vulnerabilities found by Anthropic's Claude Mythos AI model, CEO Brian Moynihan told CNBC, adding to the bank's existing 3,000-person security …
Developer Bruno Borges released ghx, a caching proxy for the GitHub CLI that returns cached read-only command responses in about 0.1s instead of roughly 1s, with a default TTL of 30s. The ghxd daemon uses singleflight co…
Claude Code's auto mode, made the default for new sessions on Pro, Max, and Team plans in August 2026, caught 89% of planted dangerous commands in Anthropic's study of 1,053 paid testers, versus 13.6% for human review — …
Developer Brandon Thomas has released three free, open-source productivity applications built with help from Anthropic's Claude AI: WordCraft, GridCraft and DeckCraft, written in Rust and aimed at Windows, macOS and Linu…