OpenAI announced an initiative to strengthen NATO and allied cyber defenses with AI. The announcement frames AI as a defensive cyber asset for allied governments.
DeepSeek's dsh-superagent-prototype (23,663 stars) and deepseek-harness (282,933 stars) were both updated on 2026-11-17, indicating ongoing work on agent harness tooling.
jingyaogong/minimind demonstrates training a 64M-parameter LLM from scratch in about two hours, trending again on GitHub. It is a reproducible, low-cost educational training pipeline.
kajisho5/ffmpeg-skill packages 42 FFmpeg tools with a machine-readable contract and MCP server for Claude Code, Cursor, and Codex, enabling local video editing without cloud or API keys. It gained 1,376 stars in a day.
elementalsouls/Claude-BugHunter is a Claude Code skill bundle for bug hunting and external red-team work, including 82 skills, 15 slash commands, and 681 disclosed-report patterns across 24 vulnerability classes.
usestrix/strix is an open-source AI penetration testing tool that finds and fixes application vulnerabilities, gaining hundreds of stars. It targets automated security testing.
The DOJ confirmed it is investigating Megaspeed in connection with Nvidia chip access, an update to a developing export-control case. The probe sits within broader US efforts to limit China's chip access.
Moonshot's Kimi-K3 GitHub repository was updated on 2026-11-17 and shows 9,095 stars, alongside continued activity on Kimi-Audio. The repo activity suggests active development on the K3 model line.
NVIDIA open-sourced SkillEvaluator, a multi-tier framework for evaluating AI agent skills and plugins with quality gates, semantic overlap detection, synthetic evaluation datasets, and live agent evaluation.
google-research released PaperOrchestra, a multi-agent framework for automated AI research paper writing, gaining over 1,100 stars in a day. It is an official Google Research repo.
Google renamed its Spark AI agent to 'Tasks' and made it free. The move positions an agentic task-execution layer as a default Google product surface rather than a paid add-on.
Google now offers a way to opt out of AI training data usage in Gemini without sacrificing chat history retention. The change decouples history availability from training consent.
A new paper, ScaleVLA, reports scaling vision-language-action models to 552 billion parameters, the largest such model reported to date. The work is gaining traction on Hugging Face Papers.
OpenAI announced GPT-6.2 Astra and Sol as its newest flagship model family, alongside a domain-specific 'Astra for Law' tuned for legal intelligence tasks. The release comes bundled with adjacent announcements on NATO cyber defense and Chile access expansion.