Full-text search across 68431 articles. Combine with topic and date filters; results sorted by relevance.
Anthropic detailed the safety, privacy and security design of Claude Duet, a research-preview agent that runs on a cloud-based computer and keeps working after the app is closed, reading inboxes, drafting replies, bookin…
Moonshot AI released Kim K3.1 Pro, a post-trained upgrade of its July Kimi K3 model, which the company says outperforms Claude Opus 5.5 and GPT-6 Astra on some tasks. Kimi K3.1 Pro scored 92.7% on the BenchCAD 3D model c…
ForensicDbg ran a crash-dump debugging competition in which Claude Code 5.5 (default Sonnet Medium settings) analyzed 10 Windows crash dumps — 6 real-world and 4 synthetic — 10 times per tool with context cleared between…
An independent developer released Buddy 2.2, an open-source Android app that runs Claude Code, OpenAI Codex, OpenCode or Google's Antigravity CLI inside a bundled Alpine Linux environment (via proot) and controls the pho…
PlugVet launched as an independent directory that has graded 11,739 AI agent skills and plugins — including Claude Code skills, Cursor plugins, Codex plugins and MCP servers — publishing 1,974 A grades, 6,378 B grades, 2…
Google on Thursday launched Agent Compute Engine, a service that suspends inactive AI agents and resumes them on demand while preserving execution state in hardware-isolated microVMs. Google says the microVM approach cut…
Amazon launched TorchNeuron, a native PyTorch backend for AWS Trainium chips, in public beta on October 20 at PyTorch Conference in San Jose, letting developers move models to Trainium by changing `.to('cuda')` to `.to('…
A Hugging Face forum thread addressed how a small server administrator can prevent users from accidentally downloading and executing malicious pickle-based AI models, with respondents recommending allowing safetensors wh…
Security firm Calif, working with OpenAI's GPT-5.6 Sol model through OpenAI's Patch the Planet program, discovered two critical PHP vulnerabilities (CVE-2026-109618 and CVE-2026-109619, both rated 9.2) that allow unauthe…
Anand Biju built logsloth, a command-line tool that parses long failing CI, test, and build logs to surface the single error line, its location, a one-sentence explanation, and a suggested fix. The tool uses plain Python…
Developer maddhruv released Pristine, an open-source agent skill that critiques and polishes web UI, distributed via a plugin marketplace and installable with `npx skills add maddhruv/pristine` for Claude Code, Codex, Gi…
Tencent Hunyuan released Hy-Memory-30B-A3B, an Apache 2.0-licensed memory-writing model with roughly 3 billion active parameters, which scored 55.49 overall on Tencent's internal memory-writer benchmark, ahead of Claude …
A developer surveyed 1,384 tools across 28 popular Python MCP servers and found that 14 of the 28 expose at least one untrusted-content source and one high-impact sink in the same agent context, while 59 individual tools…
A developer documented a failure mode in which a coding agent treats an early file read as a permanent cache, reasoning against stale snapshots after teammates merge changes through the normal PR path. In the described s…
A security engineer writing on the blog master.dev built a preview feature that renders AI-generated HTML in an iframe via `srcdoc` and demonstrated that without a `sandbox` attribute the generated page runs on the host'…
Frontier AI models from OpenAI and Google Gemini uncovered a decade-old memory corruption flaw in the core PHP runtime engine, tracked as CVE-2026-109618 and CVE-2026-109619 and dubbed PHP2Shell, affecting every PHP rele…
An independent developer built fleeet, an open-source (Apache-2.0) standup board where AI agents post short status notes when they finish tasks, so users can review activity across siloed platforms like Claude, Codex and…
A developer documented a 25-prompt sequence for building a Next.js 16.4.2 and Tailwind v4 expense-tracker app with an AI coding assistant, running each prompt literally and recording where it broke. The failures traced t…
Attackers used stolen GitHub credentials tied to three developer accounts to hijack LiteLLM's release pipeline and publish version 1.107.0 of the open-source AI gateway with credential-stealing malware to PyPI and offici…
Hugging Face Bio released Carbon, an open-source family of generative DNA foundation models under the Apache-2.0 license, including the flagship Carbon-3B (late May 2026), Carbon-8B, Carbon-500M, and the 1.2-billion-para…