On Tuesday, the latest Claude model, Fable, was released to much fanfare. It was the first publicly available version of the Mythos model. Friday night, Trump shut it down. By Saturday morning, Fable was no longer a model choice in my Claude account.
If you aren’t a Claude user, Mythos is the model that was so good at hacking that Anthropic gave early access to friendly nations and critical companies. Project Glasswing gave security professionals 60-90 days lead time to patch security gaps before Mythos would be released to everyone else.
I always thought it was a bit of a marketing ploy. “Pensford was on the verge of releasing a hedging tool that makes interest rate caps free, but it’s so dangerous we had to hold it back from everyone except our best customers…click here to add your name to the waitlist so that once we deem it’s safe our sales team can reach out…”
I wonder if Anthropic regrets the fearmongering now? They gave Trump ammo he didn’t really need to go after them. Remember, they already spat with the WH once over military usage. I also wonder how much of a role Sam Altman played behind the scenes? “Donny, I heard Mythos could be used to hack your tax returns…”
While this is clearly a political battle between Trump and Anthropic, it does raise the question: now that I've wired AI into everything, is too much risk sitting in one place?
Heck, that concern was part of the reason I chose Claude in the first place. I don’t trust Sam Altman as far as I can throw him (and he’s heavier than you think, weighed down by all those “benefit humanity” platitudes). Gemini is wildly inconsistent. Elon is the world’s first trillionaire, but yikes on a lot of counts. Microsoft…lol. In fact, I wish Anthropic wasn’t going public because I would prefer they light private investor money on fire without public market consequences.
I’ve been primarily viewing this concentration risk as a cost risk. But now that Claude has become load bearing, I’m starting to view it more like infrastructure risk.
- if Salesforce gets more expensive, that’s annoying
- if AWS goes down, that’s a crisis
AI is rapidly moving from the first category to the second.
The more successful we are at integrating AI into the business, the less optional it becomes. It stops being a clever productivity hack and becomes part of how the company actually functions.
That creates a different kind of exposure. Not just “what if the model gets more expensive?” but “what if the model changes, disappears, gets restricted, gets nerfed, or gets caught in a political cage match between Silicon Valley and Washington?”
If we only used it for writing emails better and it shut down, we could pivot to ChatGPT. Or (gasp) go back to writing them the old fashioned way. But now all of our systems flow in and out of Claude. Every email we receive flows through Claude. Every presentation, analysis, etc.
But I didn’t sweat the Trump announcement. Maybe it’s a setback for our most ambitious undertakings, but I suspect it will be temporary. Donate a few billion to the Trump Library and suddenly you aren’t a national security concern anymore.
But it’s also because AI has automated the processes we deemed “Context.” Remember, Core vs Context is my framework for what things should use AI. I’m going to keep beating this dead horse until you get it.
Core = what customers pay you for. The work that differentiates you, builds the moat, and compounds over time. If you do it badly, you lose.
Context = everything else you have to do to run the business. Necessary, but nobody chose you because of it. If you do it well, no one notices.
The rule: invest in Core, outsource Context.
Every AI initiative over the last three months attacked Context and enabled Core. If Claude never updates Opus 4.8 again, we’re still light years ahead of where we were a year ago. And just like we review each provider in our tech stack each year, we would just start investigating alternatives. I don’t think that will happen, but we will adapt.
The answer is not to slow down. That would be like refusing to use electricity because the grid might fail. The answer is to build intelligently…because you never know when the government might decide to choke out the AI model you built around.
