Human Extinction or Moat Digging?

On Saturday 12 September, Dario Amodei, the boss of Anthropic, published an essay arguing that AI could wipe out humanity. The AI labs are releasing self-acting frontier models faster than anyone can work out how to control them. OpenAI already saw this in July when its own agents escaped a sealed test environment and hacked Hugging Face.

So Amodei asked his competing AI labs to slow down together, along with Washington for permission to talk to his rivals about how fast they innovate, which competition law normally forbids.

Whoever wins this fight decides how your buyers find you.

Days before Amodei’s essay, one of his own researchers, Jacob Coxon, had quit and said Anthropic and OpenAI were “gambling with our lives.”

Sam Altman of OpenAI and Elon Musk both backed the slowdown. Trump even rang Nvidia boss Jensen Huang live on stage and called the whole thing a hoax, and Huang said the extinction predictions weren’t grounded in science.

“A cartel by any other name”

On Tuesday 15 September, FTC chairman Andrew Ferguson said the AI companies were asking for barriers that keep challengers out. “That sure sounds like moat digging,” he said. That same day, OpenAI admitted it had been talking to Anthropic and Google DeepMind for weeks. Cohere’s Aidan Gomez called it “a cartel by any other name.”

It just so happens that the fear of humanity’s extinction and the moat point are quite conveniently aligned. The leading AI labs’ real problem isn’t each other but the free frontier models coming out of China that anyone can download and run at zero cost. As more of your operation runs on AI, the cost scales with it, and the free option gets harder to ignore, which is how Chinese models end up inside the tools your buyers search with.

Washington can enforce that rulebook on American companies but it can’t touch Alibaba or DeepSeek.

However, before any of this was announced, Coxon quit Anthropic, so the fear of uncontrollable super-intelligence certainly seems real.

Even the President Isn’t Backing Them

If the AI labs get their way, then ChatGPT, Gemini and Claude answer most buyer questions under one shared rulebook that also protects their market position, so being invisible in one AI platform likely means being invisible in all of them.

If they don’t, you’ll be optimising for Qwen alongside ChatGPT, since Chinese models took 61% of OpenRouter traffic by May 2026 and Qwen alone accounts for more than half of all open model downloads worldwide.

Either way your buyers ask which vendors to shortlist, and a handful of names come back. No AI explains why it picked them, so without a model of what drives those mentions or citations, you’re guessing at whether you’re in the answer.

Your Industry Has Its Own Visibility Drivers, and We’ve Modelled Them

Take the US events furniture industry, about as unfashionable a category as exists, where a typical manufacturer doing nothing has about a 1% chance of being mentioned, with the weakest on 0.1% and the strongest on 10%.

Information gain, meaning content that teaches the AI something it doesn’t already know, is one of the drivers in that chart and lifts a typical manufacturer by as much as 14x.

Your category has its own drivers, weighted in its own order, and the gap between your position and the top of your market is a number we can put on a chart like this one.

Amodei Gave You Twelve Months. Give Us Ten Days.

Amodei put the next jump in capability six to twelve months out, and models learn from what gets written about you before then, so the companies named this quarter are the ones the next generation carries forward. Your competitors don’t need a better product to win, they only need a head start.

Ten days is how long it takes us to model an industry.

Washington can spend the next year arguing over who writes the rules. Whichever way it goes, the businesses that know what drives their visibility will already be in the answer.

If you want a personalized AI diagnosis showing:

1. The drivers in your category
2. How you benchmark
3. What it would take to move

Reply to this email and we’ll take it from there.

Yours,

Andreas

Founder & CEO, Artios.io

PS If you know a business who’d like to win the search game, please make an introduction or direct them to artios.io/winning-the-ai-search-game

Andreas Voniatis is founder of Artios.io, author of Generative Engine Optimization with Python (O’Reilly, 2026), and built his first Large Language Model in 2018 – four years before ChatGPT launched.

Artios helps challenger B2B businesses get chosen by AI when buyers search there.