Full-text search across 65780 articles. Combine with topic and date filters; results sorted by relevance.
A preprint by Jeanne Salle, Hardik Bhatnagar, Matthias Bethge, Sahar Abdelnabi and Maksym Andriushchenko (arXiv:2610.31778) found that after a coding agent's first compaction, roughly one-third of the anti-cheating rules…
Sophos and Palo Alto Networks Unit 42 both retracted claims that the PHP2Shell vulnerabilities CVE-2026-109618 and CVE-2026-109619 had been exploited, five days after the critical PHP bugs were disclosed, but researchers…
Zerostack, a lightweight coding agent written in Rust and optimized for memory footprint and performance, has shipped an update, according to the project's own listing. The project reports 1.7k stars on GitHub and a late…
Developer aiuon released DKC32X, a build tool that statically recompiles the 65816 code of Donkey Kong Country (USA, revision 1.0) into SH-2 C code so the SNES game runs on Sega 32X hardware, with Claude Opus 5.5 credite…
A new benchmark, SCI-AGENT-EXAM, finds that even the best-performing frontier LLM agent, GPT-6 Astra, scores 80.9 overall but exceeds a 0.95 success threshold in only 26.0% of its runs. The benchmark, detailed in arXiv p…
The Mixup project hand-translates console game machine code into readable, modifiable JavaScript for Batman: Return of the Joker, Gradius, and DoDonPachi DaiOuJou, explicitly avoiding emulation, with Claude Code listed a…
Forever Security researchers Gal Weizman and Omer Samet reported on October 22 that they stopped an AI coding agent from installing a malicious npm package after tracing the instruction to an 'AGENTS.md' file in a custom…
A developer released Repotruck, an open-source tool that mirrors and organizes GitHub repositories through an interactive terminal UI and a Model Context Protocol (MCP) server on stdio. The MCP server exposes workspace t…
An independent research project called gcnlle has statically recompiled the Nintendo GameCube boot ROM (IPL) and modeled hardware components including EXI, SRAM, RTC, memory-card, VI, GX, DSP/AI, DI and SI, reaching and …
A developer running a daily autonomous Claude Code agent on a single Windows PC replaced a floor-based visitor estimator with a per-product ledger of the agent's own page opens, causing the reported non-self visitor coun…
A developer released php2shell-migration, an open-source AI agent skill that helps developers detect and remediate the PHP2Shell vulnerabilities (CVE-2026-109618 and CVE-2026-109619), which allow unauthenticated remote c…
RevOS routes all reported figures through deterministic database computation while using probabilistic large language models only to propose data models, metric definitions and SQL, according to a post laying out the pla…
Meta's research organization facebookresearch published the InteractiveWorld repository on October 24, 2026, the official code for a paper titled "InteractiveWorld: A Verifiable Human-Centric Environment for Building Bet…
A developer released OpenFromSource, a Debian package management tool for modifying, building and installing packages, with all of its code generated by AI. The developer wrote that the project began as a tryout for Spec…
A developer built Coppice, a local desktop app that provides a board for managing multiple CLI coding agents such as Claude Code, Codex, Cursor and OpenCode. The tool places the user in a manager role with two human appr…
Hinoki Security released Hinoki Offensive One (HO1), which it calls the first open-access offensive cyber model, scoring 88.1% on CyberGym — 2.5 points above OpenAI's dedicated GPT-5.5-Cyber. Hinoki says HO1 works throug…
Microsoft released Microsoft Execution Containers (MXC) to general availability on October 7, 2026, offering a sandbox for AI agents. The release follows a month of agent security failures: GitSpawn let untrusted git rep…
Simon Willison released battledrones v14, adding a four-player online mode and a new map called "Compass Rose" built with Claude Opus 4.5. The release modifies the WebSockets multiplayer network code so up to four human …
A team of researchers released FusionEval, a benchmark (arXiv:2610.30162v1) for evaluating model fusion across controlled expert pools built from Qwen3 and Llama base models spanning math, code, science, instruction foll…
A developer's writeup warns that AI coding agents frequently produce unit tests that read the live system clock, so a test that passes on the day it is written can fail a week later in CI with no product change. The piec…