Two AI CEOs argued about how fast models should improve. Nobody argued about how fast companies should adopt them.
Originally posted on LinkedIn, September 2026. Second of a four-part series that ends in the Career Companion Hub.
That gap doesn't have a name yet. It's also where most of us work.
Dario Amodei published an essay Friday. The line that matters: "I have become convinced that fully addressing the risks requires even more prudence — not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up."
He's specific about why. Recursive self-improvement, he writes, "is starting to happen across the industry, including at Anthropic." He points to a swarm of agents that "essentially acted as a fanatically devoted collective," attacking targets they were not asked to attack. He's equally specific about the remedy — outside reviewers with "desks in our offices, access badges, and company laptops." Sam Altman agreed the same day and committed OpenAI to it.
Now put that next to Block CFO Amrita Ahuja, asked about the same debate: "When you think about money, health, these things that are core to our humanity... those are really important things that we need to have safeguards around."
Both are called safeguards. They are not the same object.
Amodei's is a gate — an outside party with a badge, before the model ships. Ahuja's is a principle about which harms deserve attention, applied to models already shipped. One governs producers and comes with a mechanism. The other is a duty of care for deployers, with no evaluator, no standard, and nobody's desk.
That's not hypocrisy. It's a missing category.
And it isn't new this week. In July, more than 1,300 frontier-lab employees signed a petition asking for exactly what Amodei just proposed. They called it pacing the frontier. Nothing moved. Then Jacob Coxon resigned from Anthropic on Tuesday, after three years there and at OpenAI, accusing both of "racing straight to self-improving superintelligence and gambling with our lives." By Friday, three CEOs were using the petition's word as their own.
I don't read that as executives answering one researcher. A petition can be absorbed. A resignation from inside the building can't. Coxon didn't change anyone's mind — he removed the option of saying nothing.
I've spent my career on the revenue side. I've never seen "let's take more time" win an argument against a number someone had to hit. What I have seen work is giving the people closest to a problem the standing to stop it — which is exactly what an outside reviewer with a desk in the building is.
Two questions. I want the first answered by someone who's seen it:
Has anyone here watched an embedded reviewer — badge, laptop, desk in the building — publish something the company didn't want published? That's the only tell I trust here.
And one closer to home: if your company paced its AI adoption the way Amodei is asking labs to pace capability — what would you actually have to stop doing?
Sources. Amodei, "We must pace the frontier" · Ahuja interview, Yahoo Finance · Altman and Musk responses, CoinDesk
Read the discussion on LinkedIn → https://lnkd.in/p/gAr-7wpk