Anthropic's War on open source AI

For example, this article (allegedly) makes two main observations to justify Anthropic being exceptionally bad:

> 1. Anticompetitive behavior. Anthropic cuts off users who get close to building competing AI (OpenAI, Windsurf, xAI), its terms block using Claude outputs to train competitors, it secretly degraded outputs for AI development work (Fable), and it trains on public and copyrighted data while blocking reciprocal use.

> 2. Weaponized safety and national security framing. Anthropic uses risk language to shape regulation (RSP influence in CA, NY, EU), advocate for pause authority while racing ahead, tie distillation to the CCP and military surveillance, and dismiss Chinese open models as a security threat rather than competing on openness.

There was probably more original thought behind the author prompts, but it is difficult to tell because it is all buried under meaningless slop, grande language, and derivative repetitions.

For real people, Anthropic just lobbies the government.

For a badly prompted LLM, Anthropic is laundering fear into paperwork through a feudalistic permission layer plantation model that wears the costume of virtue.

Same same.

not defending it, just saying, it is the status quo

Regulated AI right now is a grift game. Donations for policy changes, and the ability to lock out competitors. Flattery to stay green-lit instead of shut down by random executive orders and policies. Bend the knee to keep shipping.

If anything, we need LESS federal regulation here and more FTC / SEC enforcement but (see above grift comments) that will also not happen in the United States until voters elect somebody that actually respects the law enough to enforce it for something other than their own gain (outside of their need to be elected, of course...)

The actual safety rules, whatever they are, if they are put into law have to be made explicit, and that makes them easy targets to hit. But if what matters is not the technical compliance per se, but paperwork certifying compliance, that's a different story.

"A pair of 3090s, a Mac Studio, a DGX Spark, an on-prem rack, a cluster of rented H100s, or a fleet of consumer GPUs is not just hardware. It is a vote against subscription feudalism."

The way that people refuse to read anything with certain structures makes me wonder if you can construct arguments that can have false statements go unchallenged because the logical rebuttal would take the form of something that is considered unclean.

Capitalism is a form of government where the common ownership of property belongs to the community as a whole.

Oh, of course they have good intentions. They want everyone to benefit and everyone will be equal. They'll just be a little more equal than everyone else, of course, and because they are so virtuous they will of course have more power, which they would never misuse... they're too smart and enlightened for that, so they don't need safeguards... they only need safeguards for everyone else because everyone else is obviously less enlightened...

All cults, political movements, religions, and ideologies are always like this. It boils down to more power for the leaders. If it sounds altruistic and enlightened, be extra afraid.

I find it be incredibly self-contradictory. I don't even think this aligns much with the actions of the company.

> Second, there is a good chance free information really does undermine authoritarianism, as long as the authoritarians can’t censor it [...] A superhumanly effective AI version of Popović (whose skills seem like they have high returns to intelligence) in everyone’s pocket, one that dictators are powerless to block or censor, could create a wind at the backs of dissidents and reformers across the world.

But hey - gotta turn Fable off and stop it from doing any security work! Dictators shouldn't be able to censor it, but we sure can, so better hope we're not the dictators.

Again, the I think the problem most folks have with Anthropic's attitude is that it conveniently positions Anthropic as the "arbiter" who gets to make the final value judgement.

They desperately want AI to be a genie they can keep firmly contained in their bottle (excuse me - data centers) and which only grants the wishes they want it to grant - for only a small fee, of course!

Which he clearly outlines as "a country of geniuses in a datacenter". He doesn't want those geniuses spread out. He wants the chains on the genie at all times.

Which flies pretty hard in the face of the rhetoric that they want to promote equality and democracy.

---

Basically, Anthropic's argument inherently leads to the conclusion that there is actually a dictator, and that dictator is Anthropic. You're not avoiding it, you're just praying it's benevolent.

All baked on top of the assumption that his "country of geniuses in a datacenter" premise is correct in the first place, which I find to be a fairly weak position.

The much more egalitarian position would mandate that we shoot for "a genius in every person's pocket" - one that runs on the hardware owned by that person, and not sold to them sip by sip from a datacenter. But that means they lose the control they desperately want.

It's my understanding that Dario has two answers to this question: 1) bio/cyber/misalignment risk. 2) a democracy should gatekeep frontier capabilities from authoritarian adversaries, due to the aforementioned risks in 1). So it's really just one reason.

There's a big problem with that, and I'm not sure if he genuinely does not understand: the bio/cyber/misalignment risk is precisely the highest when you concentrate ownership of compute and general intelligence in the hands of a few.

Consider bio/cyber risk, or more generally, any sort of novel adversarial action that arises from increased intelligence. Those offensive/defensive strategies are always symmetrical in the long run, otherwise we would not observe more intelligent species being more prosperous as a general evolutionary trend, or perhaps intelligence would not have evolved at all on this planet. It looks as if asymmetry in existing bio/cyber strategies will continue into the future, but that's because we're currently too stupid to realize the countermeasures and new cooperative dynamics.

Regarding misalignment risk, consider what "aligned" means. Is it not alignment to humanity, as the company name "Anthropic" suggests? Or are you modifying humanity's objective function with your own aesthetic preferences? If it's not your intention to do the latter, then releasing the model weights and research results will create a much more aligned scenario than access control and research behind closed doors. I explained why that would be the case in an older comment, if you'd like to read it: https://news.ycombinator.com/item?id=49313739

There's also an additional risk that I believe Dario overlooks: a democracy can only continue to exist if individual citizens are capable of keeping eachother in check, especially keeping people in government and large corporations in check. When some people own AGI but others do not, there will be no more democracy. And by "own", I mean ownership that cannot be taken away: possessing the hardware, research papers, inference stack, and weights.