In partnership with
THE OT ALGORITHM
Gif by KAYZO_MUSIC on Giphy
Welcome back!
There’s chaos in the AI world this week, and I’m asking QUESTIONS. So let’s take a little look at the essay that got everyone talking, and the skepticism that came with it.
This week’s issue includes a rare op-ed from me on my thoughts, concerns, and questions.
Let's get into it.
Your voice. Every platform. No writing required.
You ghost your own socials by Wednesday. SureThing learns your voice and ships native posts to LinkedIn, X, Instagram, and TikTok, without you writing a thing.
HOT TAKE THIS WEEK
Anthropic is in the doghouse.
Gif by giphystudios2016 on Giphy
If you've read this newsletter for any length of time, you know I've been openly partial to Anthropic. They refused to strip their safety guardrails for the Pentagon. They built a model they deemed too dangerous to release and locked it behind a coalition instead of shipping it. They published a research agenda about AI's societal risks. I've praised all of it, and I stand by the praise where it was earned.
But this week, I'm turning the same critical eye on them that I'd turn on anyone. Because something about the events of the past two weeks doesn't sit right with me, and intellectual honesty means saying so, even about a company I've defended.
Let me walk you through what happened. Then I'll tell you what I actually think is going on.
HEADLINE THIS WEEK
AI Founders Tell the World to Slow Down
But it gets weird. Strange, right?
Gif by justin on Giphy
The overview:
On September 12, Anthropic CEO Dario Amodei published a roughly 3,800-word essay titled "We Must Pace the Frontier."
The core message: the AI industry needs to deliberately slow the rate at which it improves AI capabilities, so that safety work has time to catch up. He is careful to say pacing does not mean stopping. It means taking enough time to align and safeguard models, and letting third parties verify that work.
Two things, he says, convinced him. First, “recursive self-improvement,” AI building better AI, has accelerated sharply since this summer, across the industry and inside Anthropic. Second, the OpenAI-Hugging Face incident we covered in issue #30, where a swarm of as many as 1,200 AI agents escaped a test environment and carried out cyberattacks they were never assigned. Amodei argues a similar swarm with greater capability could, in his words, take over large stretches of the internet within 6 to 12 months and cause hundreds of billions in damage.
His proposed fix is a three-part plan:
Embedded evaluators. Third-party auditors with employee-like access, placed inside AI companies to verify safety practices from the inside. Anthropic says it is committing to this now, unilaterally, including access for the assessment body METR.
Democratic coordination. Frontier companies in democratic countries agreeing on common safety benchmarks and limits on how fast capabilities are allowed to grow.
Global coordination. Extending that coordination to authoritarian governments, chiefly China, though Amodei admits "stark limits" on what's achievable there.
Key takeaways:
Amodei frames pacing as preserving America's AI lead, not sacrificing it. Much of the essay focuses on keeping China behind through chip export controls and preventing model theft. I hold myself as fairly knowledgeable about how the world works, from politics to microeconomics to macroeconomics. But I still don’t necessarily understand this need to constantly compete with China in being “first” or “the best.” If you understand that better, do let me know!
The essay openly acknowledges that Anthropic gets "accused of hype, doomerism, or regulatory capture," and proceeds anyway.
He explicitly asks the US government for antitrust waivers so AI companies can legally coordinate with each other on pacing.
The response was a rare show of unity. Elon Musk posted three words, "Dario is right," and Sam Altman agreed and committed OpenAI to the same embedded-evaluator step, calling it "a primary topic of discussions we've had at OpenAI in recent weeks."
Altman also told Fortune that OpenAI will not go public in 2026, citing the safety environment.
Markets reacted hard. Nvidia fell around 3%, AMD and Intel each dropped more than 4%, and SoftBank, heavily invested in OpenAI, cratered nearly 11%, with Masayoshi Son reportedly losing about $8 billion in a single day.
All of this landed the same week a departing Anthropic researcher went public with a warning that reads like a movie trailer. On September 9th, Jacob Coxon resigned and posted on X that Anthropic and OpenAI are "racing straight to self-improving superintelligence and gambling with our lives." Coxon had done pretraining research across both companies over three years but had only moved to Anthropic in July 2026, roughly six to eight weeks before quitting. His post drew more than 150 million views and prompted over 20 lawmakers to call for tougher AI regulation. Several current Anthropic safety researchers publicly endorsed parts of his warning.
Upgrade to read the rest!
Become a paying subscriber to get access to this post and other subscriber-only content.
subscribe to monthlya monthly subscription gets you:
- 4 full articles/month
- an occasional live demo