For this newsletter, I try to stay in our lane: the intersection of AI, work, and how companies compete. It's wide territory, and a lot of fun to cover.
Yet this past week, dozens of you reached out asking, "What should I make of this whistleblower at Anthropic?" and “Are we all doomed?” So, for what it's worth, I'll swerve out of our lane for a moment.
Four quick thoughts:
1. The "whistleblower." The trigger was Jacob Coxon, a 27-year-old researcher whose resignation post on X blew past 150 million views in a day. But the incident looks strange. Coxon spent only a few weeks at Anthropic, in a junior role. He didn’t cite anything specific. And the amplification looks coordinated, not organic: a WSJ exclusive that ran before the post, his largely unfollowed X account picking up 200,000 followers overnight, and a cluster of other odd, perfectly-timed PR moves.
2. Anthropic's response. Whatever Coxon's motives, what matters is Anthropic's reaction. Colleagues echoed his alarm, some pegging the odds of AI catastrophe above 10%. Then, on Saturday, Dario Amodei published a 3,800-word essay urging the industry to "pace the frontier" and committing Anthropic, unilaterally, to embedded outside evaluators. Wild stuff for a company filing an S-1 (let’s just say human extinction is quite the risk factor!). And it raises the obvious question: is this regulatory capture by a market leader or legitimate concern? And if it’s the latter, how can they still go public? (Notably, later in the weekend OpenAI said it would delay its IPO over the issue.)
And what if Anthropic, OpenAI, xAI and Microsoft decide to slow things down? AI progress certainly wouldn’t stop and multiple Chinese labs would keep pushing the frontier. Then there are open source models, nearly as capable, already circulating freely and impossible to claw back once released. As such, unilateral restraint (however defined and monitored) is not a practical solution.
3. It's election season. Certain issues in American life sit dormant, then explode into consciousness the summer before a vote. You can set your clock by it. This cycle, AI (specifically with doomerism and data centers) is political red meat. Expect the hyperbole to continue to amp up. But judge where we actually stand on these issues in January, once the election dust settles.
4. You, and your company. This is the key point. Businesses today are tapping a small fraction of what existing models do. Even if the labs ship nothing new from here, companies like yours could still radically transform their products, customer experiences, cost models and overall competitive position with what's on the shelf. The constraint is not AI capability. Instead, it’s the boring stuff like change management, process redesign, user acceptance, tech integration, etc. So while arguments rage in the public square, keep running the hard yards of mapping AI to your teams, workflows, products, and P&L. That remit didn't change this week.
In summary: We have a habit of taking a genuine risk and catastrophizing it, extrapolating our fears to one worst-case scenario (think nukes, climate change, bioweapons, disease, immigration, etc.). AI just joined the list. But like the others, the armageddon scenario is only one of many, and it sits alongside the far likelier futures where AI helps cure diseases, personalizes education for every child, and compresses decades of scientific progress into years. We can dwell on the worst case until we're rendered catatonic, or we can get back to the day-to-day work of building those better outcomes. I know which one I'm choosing.
The week's most important AI news, curated for you:
