Regulation for Thee, Not for Me

Everyone knows the old proverb, "be careful what you wish for..." And while no one is sure of its exact origins, it has retained its power over perhaps centuries because while everyone knows it, many still fail to heed its warning time and time again. These past few days have brought the latest example: Anthropic.

The AI startup has been the loudest proponent of stricter controls over the broader technology, often equating it to nuclear weapons to drive the point home. Well, message received – even if garbled and mangled.

With the news late last week that the Trump administration would be putting in place export controls on Fable 5, Anthropic's latest model, the company is now once again scrambling to figure out a path forward. I say "again" because we're a mere three months from the last blow-up with the administration, which ended in the company being deemed a supply-chain risk. This new situation has nothing to do with that one, the administration is quick to say. But of course it does, even if indirectly. It's a dance between two sides that clearly don't like one another because they clearly don't see eye-to-eye on some crucial policies and ideals. This, in turn, seemingly gives the administration a decidedly quicker trigger finger when it comes to things such as this current situation.

And Anthropic is giving them every excuse in the world to do this because again, they're asking for it. Literally! Again, the endless calls for stronger controls and oversight, while perhaps noble in ways and prudent in others, are simply not a good way to impact policy. At best, you're going to be charged with fear-mongering. At worst, you're scaring away would-be customers.

Now, Anthropic may say that actually, these stances are engendering some level of trust with the customer base. The notion that they're perhaps the only ones telling the truth about AI and even willing to stand up to the government to prove it. Certainly, their recent surge in numbers plays to this narrative. But it's also far more nuanced than this. Because beyond the battles with the government, Anthropic's actions are also undermining such trust in other ways.

There's a growing sense that it's increasingly Anthropic's way or the highway. That applies to the administration and to partners and to customers. And that's perhaps okay when you're the scrappy underdog. But when you move into the position of power as the de-facto leader in AI, that's a problem. Potentially a big one.

As such, when the company complains that the export control around Fable is unwarranted, it has the decided ring of "regulation for thee, not for me." I'm not saying they don't have a point – in fact, based on everything I've read and heard, I think they do (more below)! But the point is that this misses the point because Anthropic is increasingly missing the point. The company does not operate in a perfect simulation, but rather in the real world. The very messy, very chaotic real world. The reality here, sadly, is that you must often bend to get your way. Because you have to work with others. Sometimes that involves bending to the point of breaking, the key is simply not to go over that line, obviously.

You would have thought Anthropic would have learned this lesson already. They were against taking money from certain entities in the Middle East until the reality of the world dictated that they had to in order to survive. The DoD situation was one the company seemed fine going to war over, until the realities of actual war dictated that they had to tone down the rhetoric and let cooler heads prevail.

But, well, they keep finding themselves in compromising situations! It's to the point where you almost have to wonder if Anthropic isn't actually angling for such outcomes. That is, while calling for regulation is one thing, forcing yourself to be regulated is, well, a more direct way to enact such regulation! Wait a minute. What if this is what Anthropic wants?!

Again, it simply can't be discounted to zero at this point. There's a way to tell this story where what they're doing is actually forcing the government to act and to make an example of them for the greater good. Well, for Anthropic's version of the greater good, at least. "Regulate me, Daddy," and all that.

The problem with this narrative is that the administration doesn't seem eager or ready to regulate anyone else except for Anthropic. So yeah, that's a problem if that's your game of 3D chess. It turns the game into checkers, fast.

Let's try some triangulation to Occam's razor this shit...

At some point last week, it seems like the Trump administration was alerted of a potential security problem with Fable. Multiple stories point to Amazon as the main whistleblower, though other reports have multiple entities raising such concerns. At least one report notes that Amazon flagged this to Anthropic first, and this is plausible because it sure seems like the actual issue here is that Anthropic doesn't believe this vulnerability is a big deal, while the administration does.

So it's entirely possible that Anthropic was made aware of this issue, but decided there was no immediate action needed because, in fact, the precautions they took to turn Mythos into Fable should take care of the concern raised. Further, it seems like the same vulnerability – a bypass, not a "universal jailbreak", per Anthropic – exists in other leading models and they clearly don't view it as an exploit that's going to lead to catastrophic harm. Again, this is Anthropic, the company quick to jump on any and all reasons to declare that some element of AI is going to cause potentially existential harm!

Anyway, again, the administration clearly believed otherwise. Or at least that's the cover they're conveying. It's certainly possible that someone convinced them of this, but it's also entirely possible that they're wrong. Regardless, the situation led to a phone call to Anthropic the outcome of which was... not great!

A lot of he said/she said (he sAId/she sAId) here. Clearly someone on the administration side leaked out that when they tried to get Dario Amodei on the phone, he was said to be out of office at a wellness retreat. The fact that Anthropic flat out denies this is amazing. It suggests that someone really wanted to make Anthropic look unserious in this situation. "Sorry, Dario couldn't be bothered to save the world because he was in a sound bath." That kind of thing. I can't wait to hear the real background of this aspect – it does seem like he was stuck in something , was it a meeting about wellness, something about the model and/or safety, that was distorted, intentionally or not, into a "wellness retreat"?

Whatever Amodei was busy with, the fact that he was busy when the White House called is obviously not a great look. No matter what you think about the administration, it's the White House! You should probably have some level of respect for the institution and their concerns over safety. Even if you think they're misguided! You probably shouldn't leave them waiting for a – very specific – hour and fifteen minutes. Sure, others may have been tasked with managing the situation, but again, it's the White House. This is a CEO situation.

And then when the phone calls were finally connected, it sounds like that didn't go well either. To the point where Scott Bessent and perhaps Susie Wiles – notable as they were said to be two of the key drivers to cool down the previous DoD situation – are clearly being conveyed as annoyed/pissed off. Again, the reporting suggests that Amodei gave the administration the "this is no big deal" explanation and it sure seems like they didn't like that explanation too much! One can imagine they viewed this response as flippant.

Anthropic was given an hour and a half – 90 minutes! – just fifteen minutes longer than Amodei left them hanging on the phone, to pull Fable.

Anthropic ended up doing so after the White House took the drastic action of the export ban. This was not only a problem for other countries – including allies – looking to use the model, or its variations, like Mythos, continuing their security research, but also for American companies with non-US citizens working there. Including, Anthropic! So yeah, Anthropic had little choice but to pull the model.

Again, it feels like this is all a bad game of telephone, both figuratively and literally.

It's still not entirely clear which side is "right" and there's undoubtedly nuance here. It sounds like most outside experts are siding with Anthropic right now, recognizing what the problem is and recognizing that the administration may not realize what the actual issue is. OpenAI is oddly – perhaps not so oddly – quiet here even though their own models may have the same issues. Ditto for Google, xAI, and everyone else. Many rushed to defend – or at least pretend to – Anthropic in the DoD situation, but there are a lot of crickets here. If nothing else, it showcases that Anthropic is perhaps burning more bridges than just those with the Trump administration.

At the same time, China got quickly thrown into the mix. Just as a bogeyman, or as a legitimate concern? Unclear. Anthropic, for what its worth, says China was never brought up as an actor of concern in all this...

And why on Earth is Andy Jassy, fresh off of a massive capital infusion from Amazon to Anthropic, in the middle of this? He can't be the actual whistleblower, right? Right?! Maybe he just happened to be on the phone with President Trump, or some other high-ranking official, when all this was going down? And when pushed for his thoughts, he just relayed what his researchers found? There are some great memes around this though, as you might imagine.

Again, this is all just some triangulation and guesswork to come to the Occam's razor of it all. It's likely a misunderstanding – on a few fronts. And it's undoubtedly exacerbated by the existing bad blood between Anthropic and the Trump administration. As such, Anthropic finds itself once again scrambling, not necessarily to fix Fable, but to convince the administration that they don't have to.

This all points to that bigger issue: the complete and utter lack of trust between the two sides. Arguably the most important AI player in the world and the entity that runs the free world – and the country in which said AI player operates. That seems like a bad disconnect to have.

The good news for Anthropic is that the Trump administration won't be in power forever. The bad news for Anthropic is that they will be in power for another two and a half years. A lot can and will happen in that time – especially in AI.