Frontier Labs Want to Slow Down, OpenAI Delays Its 2026 IPO, Anthropic Flags 5 Bioweapon Cases | EP #291

Uh, let's open with the most consequential AI policy essay of twenty twenty-six.

Saturday morning, Dario Amodei, CEO of Anthropic, published a thirty-eight hundred word essay titled We Must Pace the Frontier.

His argument: the industry must deliberately slow the pace of capability advancement so that safety work can keep up.

Quick context on why this matters.

This is not a pause AI letter, uh, from outsiders, right? This is the CEO of a company about to go public for two trillion dollars, and the company that just shift, uh, shipped, uh, Fable five point one, and honestly, one of the top two most powerful executives on the planet.

Dario says that two things changed his mind.

First, recursive self-improvement, in his words, quote, "Starting to happen across the industry, including at Anthropic." Second, the summer's containment failures with OpenAI hugging face swarm that we covered proved that, you know, safety measures don't hold at this pace.

So, um, his core claim, and let me quote from his essay, "We must slow the pace at which we impro-improve the capabilities of AI models.

Progress will still seem fast, and we must make wise use of the time we gain." He continues, "It's my worry that in six to twelve months, much of a swarm could be capable of taking over the entire internet with a persistent botnet, potentially causing hundreds of billions of dollars in damage." The next quote from Dario says, "Pacing does not mean halting model training or technically-- or halting technical progress, but ensuring companies take adequate time to align the safeguards of their models and for third-party evaluators to confirm this." Um, let's get into this.

He offers three steps, uh, you know, uh, unilaterally giving third-party evaluators permanent employee-level access so that they can evaluate what's going on.

Second, a critical mass of US labs adopt the same standards, enabling verified pacing through legislation or a narrow antitrust waiver.

And third, international coordination for escalating level, levels, banning bioweapon use, which he thinks is feasible, uh, and at the end-- other end to a full pause, which he thinks is not reasonable.

You know, uh, I'll, I'll go to, you know, to Alex on you this-- uh, for you on this first.

At the same time, I'm gonna show the slides of Altman, Hassabis, and Elon's tweet, uh, accepting this as, you know, uh, something that all-- we all need to do.

So, Alex, what's your take on the essay?

Yeah, I'll give you my unvarnished hot take.

I think what we saw this weekend and in the few days that followed m-may have been nothing less than a clumsy decapitation attempt against the takeoff itself, and I emphasize clumsy.

So I, I think it's-- I think at this point, knowing what I know and knowing what's been publicly reported, it's probably impossible to consider Dario's essay in isolation.

So consider the following sequence of events.

Events may or may not have started with Jacob Coxon, uh, whom we spoke about on previous episode, doing his public rage quit, maybe to get attention, maybe not, but either way, a public rage quit pointing at safety issues.

Then we get to the foreign state actors, which has now been reported from various venues, amplified Jacob's resignation letter, ostensibly for obvious nation-state competition reasons, possibly slash probably to decelerate American super intelligence.

Then, as far as I can tell, it seems like Dario takes the bait and Dario and, uh, Anthropic, again, this has been widely reported.

There's a, a broad alignment between Anthropic and effective altruism.

He takes the bait and publishes an essay pushing a very EA-ish agenda to decelerate.

Then the other lab leads, who we now know were already in discussions for weeks prior to Dario's essay with Anthropic about forming what I've publicly characterized as a cartel, amplify the discussion.