Anthropic's Heel Turn
This is a great post by Hari Raghavan, despite the fact that it's written on Xitter, which offers up an experience that's a bit like reading in an active war zone. He makes the case for why the sentiment around Anthropic, which started as the sort of lovable underdog of AI compared to OpenAI,1 is shifting, fast.
Part of it is as simple as moving from an entity playing catch-up to one in the lead. This naturally changes perceptions – just ask Apple. But a lot of it is the self-inflicted wounds Anthropic keeps giving themselves over and over again. None of it seems actually nefarious, but all of it suggests a complete and utter inability to read a room.
At first, this was the Pentagon, which, admittedly, is complicated, and perhaps especially under the current administration – an administration full of folks that flat-out don't like Anthropic, I should note. And that's in part because of their ideals, some of which stem from the term which has become the bogeyman of tech: effective altruism. The government perhaps thought they locked up that notion with Sam Bankman-Fried, but then Anthropic came roaring into prominence and placement with their superior models and Claude Code.
Anyway, a lot of the EA stuff is undoubtedly FUD at this point. It's Anthropic's actual rhetoric, as outlined in the many missives of Dario Amodei that is arguably more problematic now. It's turning the strength of the company – their idealistic nature and maximal stances on AI, from Dario on down, which has naturally built up a team of "true believers" who don't waver when, say, Mark Zuckerberg comes calling with billion-dollar poaching offers – into a liability.
Again, the real problem is that they don't seem to see the predicaments they're getting themselves into until after it's too late. Even after getting blacklisted by the Pentagon and quickly scrambling to de-escalate, they're right back at it with other escalations.
Again, I don't view any of it as nefarious. But they simply don't understand how the world is going to react when, say, they release their Fable model that secretly sabotages the abilities of others looking to use Claude for augmenting their own models. On the surface, we can all understand why they did this – they're sick of everyone from Elon Musk on down using Claude to distill to their own competitive offerings. And they undoubtedly also believe such actions can be used by bad actors to create bad models to do bad things. But it's insanity to handle this via subterfuge – certainly when there are legitimate use-cases from academics aiming to study AI or researchers aiming to quantify models, for example. But even if it's just to thwart bad actors, this is a bad look and policy when you're now viewed as the leader in the space – one trying to build a reputation around trust and safety no less!
This is all Comms 101 and Strategy 101.2 And to beat the dead horse, that's the real problem here. It would be one thing if this was a one-off situation, but Anthropic keeps shooting themselves in the foot over and over again. Because they clearly just can't see the problem with anything that they're doing. Because they obviously think that the fact that they believe they're doing what they're doing for the right reasons outweighs anything else. That's just not how the world actually works, sadly.
So while I think the public is now viewing this as a sort of classic wrestling "heel turn", where the hero becomes the villain – Harvey Dent, if you prefer the movie analogy – I actually think this is more like an Anakin Skywalker situation. That is, he turns to the Dark Side for arguably justifiable reasons: to save the life of his wife, Padme, and their unborn child.3 But he too doesn't fully think through the ramifications and can't think clearly about how such a move is likely to play out.
Next thing you know, he's slaughtering young Padawans and battling his master. HE WAS THE CHOSEN ONE. Which is how I think many viewed Anthropic. Which is why, like Obi-Wan Kenobi, we're so disappointed in the decision-making here. The good news? It's not too late for Anthropic. They just need to wake up to the realities of the real world in which they operate, fast.
M.G. Siegler
M.G. Siegler
M.G. Siegler
1 The rival company from which they were born out of, of course. Because of idealistic differences. Which adds another interesting dynamic to all of this. ↩
2 OpenAI clearly needed some help on this front too. Really, all of AI (and perhaps all of tech) does at the moment! But again, Anthropic's new front-running position makes this more problematic for them right now. As has long been a problem for OpenAI with their many stumbles! ↩
3 Spoiler – lol – twins! ↩
Member discussion