# News: September 18, 2026

> 13 curated links on AI agents, Python, security, Google Cloud, and engineering, selected by Médéric Hurier.

- Author: [Médéric Hurier (Fmind)](https://www.fmind.dev/)
- Published: 2026-09-18
- URL: https://www.fmind.dev/news/2026-09-18/

- [Toward Recursive Self-Improvement – How GLM Built Its Own Inference Infrastructure](<https://z.ai/blog/glm-built-its-inference-infrastructure>): Zhipu AI details how an Infra Agent optimized GLM-5.3-Flash inference, tripling throughput across 100,000 accelerators.
- [Qwen3.8-Omni-Flash](<https://qwen.ai/blog?id=qwen3.8-omni-flash>): Alibaba releases Qwen3.8-Omni-Flash, an end-to-end multimodal model delivering real-time audio-visual comprehension and low-latency speech interaction.
- [Why client SDK generation belongs in the open](<https://developers.googleblog.com/why-client-sdk-generation-belongs-in-the-open>): Google and Speakeasy open-source their OpenAPI suite to generate deterministic client SDKs, CLIs, and MCP servers.
- [Cloudflare security-audit-skill](<https://github.com/cloudflare/security-audit-skill>): Cloudflare open-sources a coding-agent skill for multi-phase security audits with independently verified, machine-readable findings.
- [Google ADK 2.0 workflows tutorial – Building reliable multi-agent systems](<https://www.youtube.com/watch?v=fOIGMZusCs8>): Google Cloud Tech demonstrates how to build reliable multi-agent systems using graph-based workflows and deterministic routing in ADK 2.0.
- [Building an AI Detection Engine That Understands Agent Intent](<https://www.wiz.io/blog/building-an-ai-detection-engine-for-agent-intent>): Wiz introduces an AI detection engine analyzing model I/O and reasoning traces to catch manipulated agent intent and unauthorized actions.
- [Automate workflows with custom starters and steps, third-party integrations, and webhooks in Workspace Studio](<https://workspaceupdates.googleblog.com/2026/09/automate-workflows-with-custom-starters-and-steps-third-party-integrations-and-webhooks-in-Workspace-Studio.html>): Google adds custom triggers, steps, integrations, and webhook support to Workspace Studio automations.
- [Hacking OpenAI](<https://www.hacktron.ai/blog/hacking-openai>): Hacktron AI details chaining a libheif heap overflow with SSO flaws to compromise OpenAI employee ChatGPT accounts and access internal repositories via Codex.
- [AI Infrastructure Explained (GPUs, vLLM, and LLM-D)](<https://www.youtube.com/watch?v=hBzUokVYQkI>): KodeKloud explains AI inference infrastructure from GPU memory mechanics and vLLM to Kubernetes routing with LLM-D.
- [Google Tests Paying Publishers For AI Answers Via Search Console](<https://www.searchenginejournal.com/google-tests-paying-publishers-for-ai-answers-via-search-console/589414>): Google tests a Search Console pilot compensating publishers whose content powers Gemini, AI Overviews, and AI Mode answers.
- [Did Google just kickstart the intelligence explosion?](<https://www.youtube.com/watch?v=LoLYw--s-5w>): Fireship breaks down Google DeepMind's Dream-RSI paper and debates whether simulated discovery logs qualify as recursive self-improvement.
- [Introducing Expert Intelligence in Gemini Notebook](<https://workspaceupdates.googleblog.com/2026/09/introducing-expert-intelligence-in-Gemini-Notebook.html>): Google launches Expert Intelligence in Gemini Notebook to query insights from over 100,000 published books and domain experts.
- [Notebooks in Gemini – A dedicated workspace for schools and organizations](<https://workspaceupdates.googleblog.com/2026/09/notebooks-in-gemini-dedicated-workspace-for-focused-organized-work-now-for-schools-and-organizations.html>): Google rolls out Gemini Notebooks to schools and organizations to organize course materials, projects, and conversations.