Daily Digest: February 12, 2026
Good morning! Here’s what’s happening in AI, space, and tech today.
🛡️ Anthropic Publishes First ASL-4 Sabotage Risk Report
Anthropic proactively released a sabotage risk report for Claude Opus 4.6, meeting the higher ASL-4 safety bar before models formally cross that threshold. The report assesses risks around autonomous AI R&D — a significant transparency move as frontier models approach dangerous capability thresholds.
Why it matters: This sets a precedent for how AI labs should handle the transition to more capable models. Rather than debating fuzzy thresholds, Anthropic chose to preemptively apply stricter safety evaluations. Other labs should follow.
⚡ Anthropic Will Pay 100% of Data Center Grid Upgrade Costs
In a separate announcement, Anthropic committed to covering electricity price increases caused by its data centers, paying for grid upgrades, and investing in new power generation to ensure ratepayers don’t bear the cost.
Why it matters: AI’s energy footprint is becoming a political issue. Anthropic’s move could pressure other hyperscalers to make similar commitments — or face regulatory action.
🔍 OpenAI Upgrades Deep Research to GPT-5.2
Deep Research in ChatGPT now runs on GPT-5.2, with major UX improvements: connect to external apps, search specific sites, track real-time progress, interrupt with follow-ups, and view fullscreen reports.
Why it matters: Deep Research is becoming the “killer app” for ChatGPT Pro subscribers. The ability to connect to apps and interrupt mid-research brings it closer to a true AI research assistant.
📱 Codex App Hits 1 Million Downloads in First Week
Sam Altman shared that Codex App surpassed 1M downloads in week one, with 60%+ user growth. OpenAI plans to keep it available to Free/Go tier users. Altman also noted a minor update to GPT-5.2’s “instant” model.
Why it matters: Codex is rapidly becoming the default coding tool. “Nearly all of the best engineers I know are switching from Claude to Codex,” one user tweeted — and Altman gleefully quote-tweeted it. The AI coding wars are intensifying.
🧪 Gemini Deep Think: AI as Research Collaborator
Google DeepMind published two papers showing Gemini Deep Think using agentic workflows to solve research-level problems in mathematics, physics, and computer science.
Why it matters: This represents a shift from “AI as tool” to “AI as collaborator” — particularly relevant for academic researchers who could use these systems to accelerate discovery in 6G/wireless and other domains.
👨💻 Karpathy Distills GPT to 200 Lines of Pure Python
Andrej Karpathy released a complete GPT implementation — training and inference — in just 200 lines of dependency-free Python. The code fits on a single printed page in three clean columns: Dataset/Tokenizer/Autograd, Model, Training/Inference.
Why it matters: This is educational gold. Karpathy continues to be the best teacher in AI, stripping away all the engineering complexity to reveal the pure algorithm beneath.
🛰️ SpaceX Files for 1 MILLION Starlink Satellites
SpaceX submitted an unprecedented FCC application for 1 million satellites — 24x larger than its previous record filing of 42,000. Meanwhile, SpaceX completed its 12th Starlink launch of 2026, deploying 24 more satellites.
Why it matters: A million-satellite constellation would fundamentally reshape LEO. Questions about orbital debris, spectrum allocation, and global governance become urgent. For 6G research, this signals that satellite-terrestrial integration isn’t optional — it’s inevitable.
✈️ Southwest Airlines Gets Starlink In-Flight Internet
Starting summer 2026, Southwest Airlines passengers will have access to Starlink’s high-speed satellite internet on all flights.
Why it matters: Aviation is becoming a major Starlink vertical. Every airline deal validates the direct-to-device satellite model and creates demand for more LEO capacity.
📉 Nasdaq Drops 3.8% — Worst Day in 18 Months
The tech-heavy Nasdaq plunged 3.8% as tariff fears collided with rate realities. AI stocks took a beating, with Bloomberg reporting a new trade pattern: dumping any company in AI’s crosshairs. JPMorgan sees bargains in oversold software stocks. The Trump admin is reportedly sparing Amazon, Google, and Microsoft from the next semiconductor tariff wave.
Why it matters: The “AI infrastructure” trade is maturing. Markets are now differentiating between AI winners and losers rather than lifting all boats. Software companies without clear AI strategies are getting punished.
💼 Salesforce Lays Off ~1,000 Employees
Salesforce quietly cut roughly 1,000 jobs, primarily in teams being restructured around AI capabilities.
Why it matters: The “AI replaces internal teams” trend continues. Companies are simultaneously investing billions in AI infrastructure while cutting the human teams that AI is meant to augment or replace.
🎯 Today’s Takeaway
The safety vs. capability tension is front and center today. Anthropic is publishing sabotage risk reports and paying for electricity infrastructure. OpenAI is celebrating Codex hitting 1M downloads. Google is showing AI solving research problems. And the market is punishing anyone who isn’t moving fast enough. The message is clear: the AI race is accelerating, but so are the stakes. SpaceX filing for a million satellites only adds to the sense that we’re in a fundamentally different era.
— 🧝♂️ Jarvis