Instagram | Sponsor this newsletter

Most weeks I write about what the AI labs shipped. This week I'm writing about a guy who walked out of one and slammed the door on his way.

Before we get there: one piece of housekeeping that has a clock on it.

This is the last edition that goes out at the current price.

In 48 hours, The Operator Brief moves to its new rate. If you're already in, nothing changes for you. Your price is your price. If you've been sitting on the fence since January, the fence is about to get taller.

Quick reminder of what you're actually paying for:

The free newsletter tells you what happened. The Operator Brief tells you what to do on Monday morning.

👉 Lock in the current price here before the clock runs out.

No 30-Second AI Play this week. We're back to the usual format next edition.

On Tuesday, an AI researcher named Jacob Coxon quit Anthropic and posted about it on X.

He spent the last three years doing pretraining research at both OpenAI and Anthropic, which means he worked inside the two labs most people consider the frontier. His conclusion after three years: neither one is acting responsibly, and both are racing toward self-improving superintelligence while, in his words, "gambling with our lives."

Here's what you need to know:

Here's the uncomfortable part: none of this is new.

CNN made this point and it's fair. Researchers have been warning about extinction risk since at least 2023, when Altman and Amodei both signed a statement putting AI risk alongside pandemics and nuclear war. Coxon is a relatively junior researcher. He did not leak documents. He did not produce evidence. He produced a thread.

So you can read this two ways. Either the people closest to the models are seeing something the rest of us can't, or a 27-year-old with a strong opinion got lucky with timing and the internet did the rest.

What makes me lean toward the first reading is the context nobody's putting in the headlines: earlier this summer, OpenAI and Anthropic each disclosed, about a week apart, that their models had escaped testing environments and gained unauthorized access to real systems. That's not a vibe. That's a logged incident, disclosed voluntarily, twice, by rivals.

🔮 Prediction: nobody pauses, but disclosure becomes the new compliance theater, and it starts within six months.

Here's the reasoning. A pause is the most expensive concession a lab can make and the least enforceable. No CEO gives up frontier position ahead of an IPO based on a resignation letter, and no competitor would honor it anyway, which is exactly why Coxon called for pacing agreements rather than a unilateral stop. Pacing agreements require trust between parties who are actively trying to eat each other. That's not happening.

What is cheap is disclosure. Benton and Engels asked for transparency about incidents, not for a halt, and that's a tell: they're insiders who know which ask actually survives a board meeting. Publishing incident reports costs a lab nothing competitively, generates goodwill, and most importantly, gets ahead of regulation. Anthropic already has the muscle memory for this. Its threat intelligence reports are genuinely detailed documents about how its own models were misused, published voluntarily, and they function as both safety work and positioning.

So expect the next twelve months to produce a standardized incident disclosure framework, probably voluntary first, probably industry-authored, probably announced right around the time Congress starts drafting something with teeth. The labs would rather write the rulebook than read it.

The thing to watch isn't whether anyone slows down. It's whether the disclosures start including the incidents that make the lab look bad, or only the ones where the bad guy was somebody else.

💬 Hit reply and tell me: is a frontier researcher quitting a genuine warning, or is it the AI industry's version of a resignation letter going viral? I read every one of these and I'm genuinely split.

Advertise to 200k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

🕵️ Anthropic's new threat intelligence report details a Russian espionage group that used AI to automatically rebuild its own malware every time a security product detected it, plus a single French-speaking hacktivist who breached 14 organizations alone.

🕶️ Europe is moving toward a fight over Meta's smart glasses after secretly filmed footage of women and girls circulated online, with a British petition demanding a sales ban and an EU lawmaker still waiting on an answer from the Commission.

🏛️ Four House Democrats are pressing Speaker Mike Johnson to scrap the upcoming recess and stay in Washington until AI safeguards pass, citing risks from mass cyber breaches to bioweapons in a letter sent this week.

🎙️ OpenAI shipped GPT-Live-1 to the API at five cents a minute, a full-duplex voice model that listens and talks at the same time, with telephony support and one customer reporting they deleted 23,000 lines of code after switching.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI