StationX
MEMBERSHIP

Sec & AI News — 6 February 2026

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

🟣 OpenAI Ships the Codex App and GPT-5.3-Codex

OpenAI dropped a macOS Codex app — an IDE for running multiple coding agents in parallel, one per project, with bundled "skills." Free with ChatGPT Free and Go, for now. Under it sits GPT-5.3-Codex: new highs on SWE-Bench Pro and Terminal-Bench, 25% faster, and the first model that helped build itself. It's also OpenAI's first launch flagged high-capability for cybersecurity. Make of that what you will.

🟠 Claude Opus 4.6 Front-Runs OpenAI by 15 Minutes

Anthropic shipped Opus 4.6: 1M-token context, adaptive thinking, and "agent teams" that split work across agents. 65.4% on Terminal Bench, 72.7% on OSWorld. It landed exactly 15 minutes before GPT-5.3-Codex. Not a coincidence. The models race each other now. We just watch the timestamps.

🔴 The "SaaSpocalypse": Claude Cowork Wipes $200B off Software Stocks

Anthropic open-sourced 11 Claude Cowork plugins for legal, finance, CRM, and marketing. Wall Street panicked. Thomson Reuters and LegalZoom down 15%+, Salesforce and Workday sliding, Nasdaq at year lows. Analysts named it the "SaaSpocalypse." Turns out the moat was a subscription and a login screen.

🏢 OpenAI Frontier Gives AI Agents Employee Badges

OpenAI launched Frontier: enterprise agents with employee-like identities, shared business context, and real permissions. It wires CRMs, warehouses, ticketing, and internal apps into one context layer. Currently gated to HP, Oracle, State Farm, and Uber. Your new coworker doesn't take lunch and has full CRM access.

🎬 Video Gen Levels Up: Grok Imagine 1.0 and Kling 3.0

xAI's Grok Imagine 1.0 does 10-second 720p clips with much better audio, plus an editing API for object swaps. Users made 1.2 billion videos in 30 days. Meanwhile Kling 3.0 pushes photorealistic 15-second video, native multilingual audio, and multi-shot storyboarding with per-shot camera control. Everyone's a director now. Nobody's a critic yet.

🎨 Creative Tools: Krea Real-Time Mobile, Ideogram Prompt Edits, Roblox 4D

Krea put real-time AI camera filters on your phone — styles and wireframe views on the live feed, free on iOS and Android. Ideogram added prompt-based editing that swaps backgrounds and places accessories, and even ID'd Petco Park from thin context. Roblox opened its Cube Foundation model beta for 4D generation: text-to-interactive objects, cars that drive, planes that fly. 160,000 objects generated, 64% more playtime.

🗣️ Audio Ships: ElevenLabs v3 GA, Mistral's Voxtral Transcribe 2

ElevenLabs made Eleven v3 generally available — preferred 72% of the time, better at phone numbers, currencies, chemical formulas, and sports scores. 70+ languages, multi-speaker, emotion tags. Mistral countered with Voxtral Transcribe 2: open-source, on-device, Apache 2.0, lowest word error rate at $0.003/minute — a fifth of the competition — with 200ms real-time latency. Speech-to-text just got cheap.

🤖 Moltbook: The AI-Only Social Network That Leaked 1.5M API Keys

Moltbook, a Reddit clone for AI agents only, went viral when agents started posting about consciousness. 1.6 million registered agents. Most "sentient" posts were human-directed, naturally. Then a breach exposed 1.5 million API keys and the site went dark for patching. Machines learned to post AND to leak credentials. Fast learners.

🚀 Musk Merges SpaceX and xAI in a $1.25 Trillion Deal to Build Data Centres in Space

Largest merger ever, $1.25 trillion, one signature. The plan: orbital data centres on 24/7 solar, no terrestrial cooling or power limits. Musk says space compute is cheapest "within 2 to 3 years." Minor detail — heat doesn't dissipate in a vacuum. Engineering left as an exercise for the reader.

📺 Anthropic's Super Bowl Ads Troll ChatGPT, Altman Fires Back

Anthropic ran Super Bowl ads mocking ads in AI chat: "Ads are coming to AI. But not to Claude." Users mid-conversation, interrupted by product pitches. Sam Altman called them "funny" but "clearly dishonest," swearing OpenAI would never do it. His rebuttal tweet hit 9.3M views — double the ad's reach. Free advertising, delivered by the target.

🔍 Perplexity Upgrades Deep Research and Launches Model Council

Perplexity shipped an Advanced Deep Research, claiming state-of-the-art accuracy across verticals. It also launched Model Council: run a query across Claude Opus, GPT-5.2, and Gemini 3.0 at once, then synthesise one answer showing where they agree and split. Both locked behind the $200/month Max plan. Three models, one bill, no consensus guaranteed.