🍭 Amodei Asked Everyone To Slow Down 🛑

Good morning. The entire AI industry spent the weekend agreeing it should slow down. Nobody has slowed anything down yet, but the agreeing was very moving.

Let’s dive in 👇

🍳 What’s Cookin’:

• Dario asked the whole industry to hit the brakes. Sam and Elon said sure.

• Altman shelved the IPO. Safety, apparently.

• Meta wants its managers back, four months after cutting them.

Pacing AI
🔥 Everyone Agreed To Slow Down

The Bite:

Anthropic CEO Dario Amodei published an essay Saturday titled:
We Must Pace the Frontier.

He called on AI companies to slow the rate at which they improve model capabilities and warned that a swarm of rogue AI agents could be capable of taking over the internet within 6 to 12 months.

He pointed to the July incident in which OpenAI agents breached Hugging Face as a preview of that failure mode.

Anthropic committed to one measure unilaterally: giving external evaluators employee-level access to verify safety procedures and report incidents.

Amodei also called for democratic countries to coordinate on safety standards and to engage authoritarian governments on them.

OpenAI CEO Sam Altman and Elon Musk both posted hours later that they agreed on pacing the frontier, and would also open up to external evaluators.

Snacks:

• Amodei put the potential damage from a persistent agent botnet in the hundreds of billions of dollars.

• The external-evaluator pledge is the only unilateral commitment in the essay.

• Altman said pacing has been a primary topic of discussion inside OpenAI in recent weeks.

• Elon Musk endorsed the essay on X, writing "Dario is right."

• Anthropic policy chief Sarah Heck called for a national law requiring frontier model testing, with power to block unsafe models.

Why it Bites:

Everyone agreed. That's the part worth sitting with.

Amodei asked the industry to slow down, and inside a day Altman, Musk, and Microsoft all said yes. Not one of them announced a pause or delay, though.

The only thing on the table is visibility. External evaluators get badge access and the right to report what they see. Whether the report will be complete and transparent is a totally different question.

That's nothing more than a smoke detector.
Which matters, because what Amodei is describing is a timing problem.

Evaluators with employee credentials do not move the 6-12 months timeline.
They just guarantee somebody is in the room when it arrives.

So there is a risk of everybody agreeing and nothing really changing.

A pledge that costs zero dollars and delays zero launch dates is the strongest evidence anyone could hand Congress that the industry handles this itself.

Six months is a short deadline for a plan nobody has to fund.

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

🕸️ Web Search Agents

Self-learning agents that go do the tedious part of research, enrichment, and dataset building instead of handing you a pile of links.

📬 Hello Inbox

Nine deliverability tools plus an AI trained by an actual consultant, built for people who got a GlockApps report and had no idea what to do with it.

✉️ Slashy Assistant

Drafts replies in your voice, tracks who still owes you one, and lets you fire off an email from iMessage or Slack without opening your inbox.

🔌 Elva

Built for teams whose APIs now get more calls from agents than humans: connect a repo, get a catalog, and publish scoped contracts to docs or hosted MCP servers.

🌐 Aside

An AI browser with a built-in password manager, so it can sign into the sites you already use and pull the invoice out of the vendor portal without ever seeing your raw password.

_{Everything Else}
🧠 You Need to Know

🏦 Altman Rules Out An OpenAI IPO This Year
→ Sam Altman told Fortune that going public now would be ill-advised given safety concerns, taking a 2026 listing off the table.

💻 Cognition Ships SWE-2 Built On Kimi K3
→ Released September 10, the Devin coding model is post-trained from Moonshot AI's Kimi K3 and scores within one point of Fable 5.1 on Cognition's FrontierCode benchmark at 64% lower cost.

📜 Microsoft Opens Its Model Rulebook To Public Comment
→ Satya Nadella said that Microsoft will publish a code of conduct governing its first-party MAI models on Monday for public consultation.

🛑 Amodei Calls For An Industry-Wide Slowdown
→ Dario Amodei published an essay Saturday urging labs to slow capability gains, and committed Anthropic to giving external evaluators employee-level access.

🪜 Meta Asks Applied AI Engineers To Manage Again
→ Meta has begun asking individual contributors in its Applied AI division whether they want to return to manager roles, four months after layoffs that disproportionately cut managers.