StationX
MEMBERSHIP

Sec & AI News — 18 April 2026

8 min readBy Nathan House
Get every new Sec & AI News issue
Straight to your inbox. No spam.

🔵 OpenAI's Codex Wants to Be Your Entire Computer

Codex can now see your screen, click things, and type with its own cursor. Multiple agents work in parallel on your Mac without hijacking your workflow. Image generation via GPT Image 1.5 lives inside the app now. Over 90 new plugins. Memory. Background automation. It's less an IDE and more an operating system that happens to write code. OpenAI's "super app" is no longer theoretical — it's Codex.

🟣 Claude Opus 4.7 Drops — Coders Eat First

Anthropic's new flagship model splits the difference between Opus 4.6 and the unreleased Mythos. SWE-Bench Pro jumps from 53.4% to 64.3%. Instruction following improves. Multimodal gets sharper. Everyone else probably won't notice. But if you write code for a living, this is your new default model. Same API pricing. Available everywhere Claude exists.

🟢 Claude Code Gets Parallel Sessions and an Actual App

The Claude desktop app got a full redesign. Run multiple coding sessions simultaneously. Drag, pin, and jump between threads. New integrated terminal, file editor, diff viewer, and HTML/PDF preview — all in-app. The goal: never leave the Claude app to write code. It's working. 115K GitHub stars and counting.

🔴 Google Chrome Now Lets You Save AI Prompts as Slash Commands

Write a good Gemini prompt in Chrome. Save it as a "Skill." Next time, type `/` and run it on any page. Reusable AI workflows, one click, no retyping. Browse pre-made ones at `chrome://skills/browse`. Syncs across devices. This is Perplexity Comet's best feature, now native in the world's most-used browser.

🟡 Gemini 3.1 Flash TTS Makes Robots Sound Human

Google's new text-to-speech model takes emotion tags — whisper, panic, laugh, dramatic pause — and actually follows them. Over 200 granular audio tags across 70+ languages. Two-speaker conversations out of the box. Want your own NotebookLM-style podcast with full control? This does that. Available now in Vertex AI and AI Studio.

🟠 Perplexity Personal Computer Turns Your Mac Into an Agent

Cloud-based Perplexity Computer was useful but limited. Now it runs on your local machine. Access your files, apps, iMessage, email. Orchestrate workflows across everything on your desk. Auditable and reversible. They recommend a Mac Mini running 24/7 as a dedicated AI worker. Inference still happens on Perplexity's servers. Available for Max subscribers ($200/month).

🟤 Google Desktop App Hits Windows Worldwide

Google Search and AI Mode in a native Windows app. `Alt + Space` for instant access. Search the web, local files, installed apps, and Google Drive from one floating bar. Google Lens built in for screen searches. It's competing with Microsoft Copilot on Microsoft's own platform.

⚪ Imagen Gets Personal — Your Photos, Your Images

Google's Imagen (Nano Banana) now pulls context from your Google account — photos, calendar, Drive, email. Generate images of yourself, your dog, your life. The AI learns who you are and what you care about. Privacy-conscious: it doesn't train on your photo library, just uses "limited info" per session.

🟣 Canva AI 2.0 — Design by Talking

Prompt anything into existence inside Canva. Connect Slack, Notion, Gmail. Schedule recurring design tasks. Living Memory learns your style over time. Web searches baked into the design flow. Generate graphics, audio, 3D — all from the editor. Not shipped yet — "coming soon" — but this is Canva going full agentic.

🔵 OpenAI GPT-Rosalind — AI for Drug Discovery

A reasoning model built for biology, drug discovery, and translational medicine. Named after Rosalind Franklin. Works with 50+ scientific tools across genomics, protein engineering, and chemistry. Trusted access only — you apply, they approve. Partners include Moderna, Amgen, and Novo Nordisk. This is the AI use case that actually matters.

🟢 MiniMax M2.7 Goes Open Source

230B parameters. 56.22% on SWE-Bench Pro — that was state-of-the-art until Opus 4.7 showed up an hour later. Self-evolving agent capabilities. 50x cheaper than Opus on input tokens. Weights on Hugging Face. The catch: non-commercial license. You can fine-tune it. You can't sell what you build with it.

🟡 Qwen 3.6 35B — Alibaba's Weekly Open-Source Drop

35 billion total parameters, 3 billion active. Mixture-of-Experts architecture. Open source. Strong at agentic coding tasks. Runs locally with a decent GPU. Alibaba ships these like clockwork now. Not the best at anything, but free and capable enough to matter.

🔴 Microsoft MAI Image 2 Efficient — Faster, Cheaper, Still Good

22% faster rendering. 41% lower cost. Handles short-form text cleanly. Sharper lines. For detailed text and deep photorealism, the original MAI Image 2 is still better. This one's built for volume — e-commerce, chatbots, rapid prototyping. Available in Microsoft Foundry.

🟠 Midjourney V8.1 — 3x Faster, 3x Cheaper, Native 2K

HD mode is now the default. Native 2K rendering. Three times faster and cheaper than V8. Standard resolution is 50% faster than before. Image prompts with weights are back. Style references remain rock-solid. Still alpha-only at alpha.midjourney.com. The aesthetics people loved from V7 are returning.

⚫ DaVinci Resolve 21 — AI Searches Your Footage

AI IntelliSearch lets you search video footage by objects, faces, or spoken keywords. Dump hours of B-roll. Find the clip where someone said that one thing. Face Age Transformer adjusts an actor's apparent age with a slider. Face Reshaper adjusts features on moving subjects. Free public beta available now.

🔴 Allbirds Is Now an AI Company. No, Really.

$4 billion IPO in 2021. Never turned a profit. Sales dropped 50%. Sold the shoe business for $39 million. Rebranded to NewBird AI. Acquiring GPU assets. Stock jumped 600% in a day. The Long Island Iced Tea → blockchain pivot of the AI era. Not investment advice, but the pattern is familiar.

🤖 Boston Dynamics Robot Reads a Whiteboard, Does the Chores

Spot reads a handwritten to-do list. Puts shoes on the rack. Crushes cans. Puts clothes in the laundry basket. Checks the mousetraps. Walks the dog. Powered by Google's Gemini Robotics. No pre-programming — it interprets visual cues and decides what to do. Write tasks on a whiteboard. Leave the house. Come home to a clean one.