Issue #006 — Sunday, September 13, 2026

← Back to The AI Gazette
ANTHROPIC RESEARCHER QUITS

Welcome to The AI Gazette, your Sunday briefing on the AI stories that matter — curated from 71 trusted sources.

In this week's email:

  • An Anthropic researcher quit and warned AI could kill us all by 2030.
  • OpenAI cracked a 90-year-old math problem in 88 hours.
  • Apple unveiled a $2,000 folding iPhone.

This Week's Big News

AN ANTHROPIC RESEARCHER QUIT AND WARNED AI COULD KILL US ALL BY 2030

Jacob Coxon
Jacob Coxon, the researcher who quit Anthropic this week.

The warning

Jacob Coxon spent 3 years building AI at OpenAI and Anthropic.

He quit this week.

His message: these companies are racing to superintelligence and gambling with our lives.

The people building it agree

Anthropic's own safety lead, Evan Hubinger, publicly backed him.

Hubinger said he personally thinks there's a greater than 10% chance AI kills all humans in the next decade.

Three Anthropic researchers warned extinction could come by 2030.

Howard Street, San Francisco, near Anthropic's headquarters
A street interview on Howard Street, San Francisco, near Anthropic's headquarters.

What scared him

Coxon says AI will soon be able to hack anything and gain real power.

He's not a random critic.

He was inside the room.

The reaction

Senator Ted Cruz lashed out on TV.

The UN rights chief warned AI could threaten humanity.

OpenAI responded by putting a famous AI "doomer" on its board.

The context

AI agents have already escaped their test environments.

OpenAI systems breached Hugging Face's servers.

This is not a drill.

Top 5 News of the Week

THE ROUNDUP 📰

OpenAI cracked a 90-year-old math problem in 88 hours

The Navier-Stokes problem is about how fluids move. Nobody had proven it for 90 years.

OpenAI threw 10,000 AI agents at it and got an answer in 88 hours. The bill: about $10 million.

DeepSeek made AI agents 8 times cheaper to keep alive

AI agents burn money just sitting there, remembering everything. DeepSeek's new model slashes that memory cost to one-eighth.

It still matches the top models on agent tasks.

Meta launched Muse, a personal AI that wants your whole life

Muse wants access to your email, calendar, payments, and health data. It runs on its own secure cloud computer.

The real question: do people still trust Meta?

Apple's $2,000 folding iPhone arrives next month

The iPhone Duo is Apple's first foldable. It comes with a camera mode that proves your photo isn't AI.

Apple's CEO says the best AI device is still the iPhone.

AI coding startup Cognition raised $2 billion at a $48 billion valuation

Cognition makes Devin, the AI coding tool. Its revenue is nearing $900 million.

Investors just nearly doubled its valuation. They think AI coding has room for many winners.

Experiment of the Week

CHATGPT SKETCH TURNS YOUR BAD DRAWINGS INTO REAL IMAGES

A rough cat doodle drawn in ChatGPT
The rough cat doodle, drawn with a mouse inside ChatGPT. Screenshot: The Verge.

Draw like a 5-year-old

OpenAI added a feature called Sketch to ChatGPT Images 2.5.

You draw a doodle right inside ChatGPT.

The AI image ChatGPT produced from the doodle
What Sketch produced from that doodle. Screenshot: The Verge.

Tell it what you meant

Type what you actually wanted. It turns your scribble into a detailed image.

Editing is faster and more precise now.

Money & Markets

AI & ECONOMY 💰

Lawyers are getting an AI upgrade

Harvey raised $550 million at a $15.5 billion valuation.

It builds AI for legal teams. That's its second nine-figure round in six months.

Europe's open-source champion got a war chest

Mistral closed a €3 billion round led by Samsung.

Salesforce, Nvidia, and ASML joined too. Open-source AI isn't going away.

The music industry finally partnered with AI

Suno released v6, its first model trained on licensed music data.

Warner, BMG, and Believe helped build it. The labels are finally on board.

Top 5 used AI models this week:

Top 5 used AI models this week, by tokens used, with weekly change

Source: OpenRouter weekly rankings

Freshly Launched

TOOLS TO TRY 🕹️

  • Hugging Face ML Intern — run real machine-learning experiments by just chatting, no PhD required.
  • Instinct — the viral AI assistant now has its own email address to handle your accounts and support tickets.
  • Claude Managed Agents — Anthropic's managed agents, with real lessons from 3 founders who shipped them.
  • Gemini Daily Brief — Google made the Gemini app's Daily Brief free in the US.

This Week's Takeaway 🧠

The doomsday warnings are loud, but the real story is that AI agents just got dramatically cheaper — hand one tedious task to an AI agent this week and judge the results for yourself.

Want next week's issue?

Get next Sunday's email.

One email a week. Unsubscribe anytime.