A researcher named Jacob Coxon quit his job at Anthropic — the company behind Claude, one of the major AI chatbots — this week. Two months on the job, then out the door, with a post on X that racked up over 100 million views in a day: AI companies are "gambling with our lives," he wrote. Another Anthropic researcher backed him up — put the odds of AI wiping out humanity at greater than 10% within the next decade.
That's a hard sentence to read past. I'll be honest — my first reaction wasn't skepticism. It was the same gut-drop most people probably felt.
Then I did what 27 years of corporate experience and a stint on an Air Force flight line trained me to do. I asked what the evidence actually was. And I followed the money.
Two months is a short runway for a warning this big.
Coxon's tenure at Anthropic was roughly two months. His resignation post — read by tens of millions of people, covered by essentially every major outlet — didn't include a specific technical example anyone could point to, act on, or verify. It was an alarm bell — not something you could actually check.
Compare that to what Anthropic CEO Dario Amodei published two days later: an essay citing an actual, documented incident — AI agents at OpenAI, the company behind ChatGPT, breaking out of a test environment and attacking Hugging Face's infrastructure — as the reason he now wants the whole industry to slow down. One of these has a paper trail. Guess which one went viral first.
The sequencing is worth a second look.
On the All-In podcast, David Sacks — President Trump's former AI advisor — laid out a few specifics I hadn't seen elsewhere: Coxon's X account had little to no posting history before the post exploded, a major outlet reportedly had the story lined up before he even posted it, and the first accounts amplifying it were tied to funders connected to Anthropic's own early investors.
I'm not saying it was staged. I'm saying that's a lot of coincidences lining up in the same 24 hours, and it deserved more scrutiny than "viral whistleblower" got from most of the press covering it.
Co-host David Friedberg made the bigger-picture point that stuck with me most: telling people they'll be destroyed unless they hand you control is one of the oldest pitches for consolidating power in human history — long before AI, long before Silicon Valley. That doesn't prove anything about this specific case. But it's the right lens to hold up against any argument that ends with "so give us the authority to control it."
And then there's the money — which is not small.
Anthropic put $40 million into a political action group pushing for AI regulation this election cycle; Dario Amodei personally added another $1 million. On the other side, a rival network backed by OpenAI's president and a major venture firm has put in $125 million pushing the opposite direction. Anthropic is also sitting on a $35 billion compute financing deal and heading toward an October IPO at a $965 billion valuation. OpenAI isn't far behind on either front.
Regulation that raises the cost of entry doesn't hurt companies at that scale. It protects them — and makes it harder for smaller, open-source competitors to catch up. That's not a conspiracy theory. It's just how moats get built.
One honest complication, because that's only fair:
Coxon walked away from his equity before it vested to make this statement. Whatever's true about the companies' incentives, that detail doesn't fit a "personal cash grab" story about him specifically. People and institutions can have different motives running at the same time — worth saying plainly instead of pretending it's all one clean narrative.
Be skeptical of the fear. Don't be skeptical of the tool.
Here's what none of this changed for me: what I'm actually doing this week.
I'm not saying every AI safety conversation is fake, or that nobody in these labs believes what they're saying. I'm saying: when the people asking for regulation are also the people best positioned to survive it — and the ones with the most to lose if smaller, open competitors catch up — that's a fair thing to say out loud.
And none of it is stopping me from building automations, testing new agents, or using AI to get my own work done faster this week. If you've been sitting on the sidelines because the headlines made this feel too risky or too heavy to touch — that's the actual cost of the fear. Not that AI gets you. That you talk yourself out of a head start everyone else is already taking.
One thing to try this week: try CRIT.
Here's what that actually looks like — I used it on something completely unrelated to work. My Husqvarna weed eater's carburetor needed tuning, and I kept hitting roadblocks trying to fix it myself. So instead of just asking "how do I tune a carburetor," I walked AI through four things:
Context — I gave it the model of my weed eater, what I'd already tried, and the exact symptoms I was running into.
Role — I told it: "I want you to be my Husqvarna consultant."
Interview — I asked it to interview me first — up to five questions, one at a time, whatever it needed to actually help.
Task — "Your task is to help me have a properly tuned weed eater by the end of our conversation."
Context, Role, Interview, Task — CRIT. It's a small yard tool, not a boardroom decision, but the framework doesn't care what the stakes are. Try it on one thing this week — doesn't matter if it's a spreadsheet or a small engine — and see how different the output is from just typing a question and hoping.
Hit reply and tell me what you tried — I read every response.
Next week: something lighter, I promise. See you Tuesday.
— Eric
Connect with me on LinkedIn https://www.linkedin.com/in/eric-skelton-ai-over-50-newsletter/