Govern Your Data Before Someone Else Does

Good morning, and happy Sunday. I haven’t had the time to prioritize writing in a few weeks, but the events of the last few days regarding Anthropic and the U.S. government are too juicy to ignore.

Personally, I was a little disappointed on Friday night when I sat down at my desk because I had been looking forward to spending the night with Fable. I had worked with it all day and absolutely loved it. It was, by far, the smartest, most capable model I've worked with.

I’d spent Friday morning working with Fable on a strategy deck for an enterprise partner, pulling a handful of scattered versions into something I was actually happy with. The deck itself isn’t the point. The point is that I’d leaned on Fable hard, across real work, all day long, and it had more than earned my trust.

But anyway, I digress, because what I really want to talk about this morning is the situation that unfolded on Friday and into the night regarding the U.S. government and Anthropic.

Read on if you're interested...

So Friday night around 9pm, I don't remember for sure but I think it was around then because my daughter was asleep and I was geared up to test Fable on another client project, a front end we're building on top of Knowledge Spaces for a custom vertical, I sat down and started my first prompt.

As a little more context, because context is everything with LLMs and AI, some of you know that I run a fairly sophisticated "brain" that keeps context, rules and memory in sync across a number of agents. Obviously I run Sprinklenet's Knowledge Spaces, which is the enterprise version of what I do in my own operating system. Spaces is more geared toward the multi-user, business version of what I do for myself, because for me I don't need my own RBAC, audit logs, or easy user interfaces. I run hot on the steel, as they say, as close to the edge as possible.

So anyway, I hit enter on the context prompt giving Fable the background and objectives of what I wanted it to review and work on with me, and I got an error message. At first I thought it had something to do with the advisor setting in Claude Code on terminal. Some of you may have run into this yourselves; you can't use a model that is higher capacity than your advisor model, so I checked the advisor, thinking maybe it was still set at Opus 4.8 and if the grunt was Fable then I'd have to change it. But I quickly realized that wasn't the case. Then I jumped on X and saw what was going on.

I was a little pissed because it interrupted my Friday night plans. But having watched the drama all weekend, and shrugging off the fact that I had to go back to Opus 4.8 and my dear friend Codex 5.5, both admirable and reliable assistants, I realized that we, the world, are at a major turning point right now with AI: the politics of power, money, and how each of us and enterprises will control (or not) our own sovereign destiny when it comes to owning our information and controlling AI instead of AI controlling us.

For those of you that may not be following, Anthropic issued this statement Friday night:

And that was that. No more Fable. Fable was supposed to be the slightly stripped down version of Mythos. I guess the deal was Fable was supposed to have some safeguards that would prevent it from being used in cyber attacks.

Anyway, I have been thinking about this all weekend and how this situation is exactly what I have been talking about for close to 2 years: in the future (or right now, as the case may be) every business and enterprise needs to have a multi-LLM, multi-partner strategy for managing data. When we first started building custom RAGs in the cloud we were using OpenAI models, and eventually we started building Knowledge Spaces because we saw demand for more and more governance and options in terms of integrating data from legacy systems and creating flexible, yet highly governed by the enterprise, output options for interactive use cases with employees and customers.

I don't want to sound like a know-it-all, but this is exactly why you need middleware. You cannot hitch your wagon fullscale to a single LLM company. It's too dangerous. With the stroke of a key or the stroke of a pen, if you're entirely relying on the APIs or infrastructure from one foundational company, you're at substantial risk.

I'm actually finding a lot of people are at even greater risk than their reliance on a single foundation company for LLM use: they're now relying on the LLMs to be doing things like managing decisions on what data is accessed and how, for generating outputs. My take is that you're better off using a governance tool or layer that uses clearly defined paths and rules on what data gets accessed, when, by whom, and how it gets folded into an LLM, but ideally strictly for the purpose of generating whatever the task at hand or immediate answer requires.

Meaning, I don't think most businesses or even government organizations should be relying on LLMs as much as they do. Use them for reasoning, use them for generating the artifacts you need or the decision-support you need, but you really need to be thinking about how to abstract the decision-making process itself and how your data is governed.

As of last night I started seeing a lot of senior people and Silicon Valley kingpins suggesting that Amodei may not survive as CEO by the end of this week.

Here's David Sacks on his version of what happened:

And for the full play-by-play, Politico has a good account of how this came together: Inside the whirlwind 24 hours that led the White House to slap export controls on Anthropic…

I will be surprised if Amodei remains at the helm. Obviously the guy must be super smart and genius level to have led Anthropic this far and created the fastest growing company in the history of the world. Obviously he's no fool, and clearly he's driven by a sense of purpose that, if genuine, is admirable. I don't know enough details to comment too much on whether he's right or wrong, technically speaking.

But I do think that in the wake of Anthropic's issues with the Department of War, it was a really bad business decision to take an aggressive, non-cooperative stance with the U.S. government. It sounds to me like numerous senior officials reached out to him and tried to figure out a plan to cooperate fast. And reading that he was at a wellness retreat and hard to get ahold of, well, that conjures up images of some aloof CEO disconnected from the realities of the politics of running such a behemoth of a company.

I'm not an investor in Anthropic. I have no skin in the game of any of this other than Knowledge Spaces being the best it can be, and of course for our clients I'm always trying to figure out the optimal, fastest, most effective and secure plans and paths. But if I were an investor in Anthropic, I would not be defending Amodei right now. Whether he's right or wrong, picking the fight with the government was stupid.

I suspect some of my readers might disagree and you have valid points, I'm sure, but at the end of the day, to me it smells like a fight that could have been walked back and de-escalated. The problem for Amodei and Anthropic now is that they're developing a reputation as a company that will be hard for the government to trust and hard to work with. This is going to dog them for years unless this is taken care of, and I simply don't see how Amodei will ever be trusted again by current senior officials.

Once you cross the Rubicon, there's no going back.

So, I don't know, we'll see what happens, but I think there's a shakeup coming.

There’s broad agreement, on both sides of the aisle, that the world is shifting toward more nation-centric thinking. The fully globalized model that defined the last few decades is under real strain, and you can see countries everywhere reasserting control over their economies, their data, and increasingly their technology.

Whatever your politics, that shift is happening, and AI sits right at the center of it.

My point isn’t a political one. It’s this: the situation with Anthropic, Amodei and the U.S. government really does loop back to the greater geopolitical dynamics at play. AI is such an important layer in the future of national security, for every nation, that no company, no matter how big you are, will be allowed to thumb its nose at the largest, most powerful country in the world.

Does anyone think that China would allow such behavior? What happened to Jack Ma? I'm not saying Jack did anything bad or deserved to be disappeared and rehabilitated to the "values" he's supposed to espouse, but I'm saying it's foolish for anyone to think they could get away with it. I don't care which party is in power, this notion that the technology titans will do whatever they want irrespective of any given administration's demands is not going to be good for business. I realize the irony of suggesting that in a nationalist, capitalist society we don't want the government overly playing their hand in free markets. Not at all. But AI is moving so fast and is so powerful that there's so much at stake here.

So anyway, back to us plebeians tasked with safeguarding our own enterprise data, improving efficiency, cutting costs and embracing practical technology that drives serious ROI: I think the multi-LLM, agnostic approach is even more relevant than ever. If there was any doubt in anyone's mind that you'd be fine building your entire AI strategy and partnerships around one foundational titan, it should be crystal clear now that that thinking doesn't hold water.

Anyway, time for a late breakfast and gearing up for an exciting week.

Ok, well, this edition is more like a note from me, so maybe I'll just leave us with this. This week’s's Spotify playlist (see below) has one song, as I often enjoy putting a song on repeat and writing in one sitting. It helps my brain to just zone in on what I need to zone in on. And the song is the Spanish version of Enrique Iglesias' "Escape," or en español, "Escapar."

My thinking is there's no escaping AI. There's no escaping the battles that are coming. There's no escaping the massive change that is underway right now. There will be major winners, major losers, and the future of nation-states hangs in the balance. Some will approach governance from a place of love and principles; some will approach it from pure money and control.

I don't really know how it all shakes out, but I know we're in for a helluva ride.

One song this week. On repeat. 🔁

Enrique Iglesias, "Escapar" (the Spanish version of "Escape").
Put it on, zone in, and get your work done. 🎧

🎺🎧 Note: Web Edition Only