AI’s Neofeudal Turn: The Closing of the Digital Frontier

The author argues that restricting frontier AI models to elite corporations ends the era of permissionless digital innovation and creates a new 'neofeudal' class divide. This gatekeeping stifles independent safety research and concentrates the world's most powerful creative force in the hands of a few unaccountable labs. To preserve individual agency, the author calls for transparent access, due process, and the eventual democratization of intelligence through open-source alternatives.

Key Points

The transition from a permissionless internet to a 'neofeudal' AI landscape where capital-heavy incumbents hold a monopoly on superhuman intelligence.
The failure of the 'Manhattan Project' analogy, as intelligence is an economically valuable creative force that will inevitably be pursued globally regardless of local restrictions.
The irony that restricting frontier models prevents independent safety researchers from conducting rigorous experiments on the most capable systems.
The need for AI labs to implement 'due process' and transparent appeals mechanisms for model access, treating intelligence as a public utility rather than a private gift.
The historical pattern of 'value extraction' where labs train on public data but concentrate the returns within a small circle of well-connected partners.

Sentiment

The community is predominantly skeptical of the article's thesis. Most commenters reject the 'neofeudal' framing as melodramatic, viewing Anthropic's restrictions as either temporary marketing, a compute limitation, or a reasonable security precaution. While there is sympathy for concerns about access inequality and corporate gatekeeping, the majority believe open-source alternatives and market competition will prevent a permanent two-tier system from forming.

In Agreement

Restricting frontier models to elite corporations while the public gets inferior tools creates a two-tier system that undermines the democratizing promise of the internet
Companies like Anthropic have no accountability mechanisms comparable to governments yet wield quasi-governmental power over access to intelligence
The era of permissionless AI access (2019-2025) may be a brief historical blip as companies realize more profit from internal use than from selling API access
Depending deeply on privately-owned AI systems gives those corporations outsized control over businesses that rely on them, with the power to raise prices or revoke access unilaterally
Once AI capabilities are gated behind credentials and enterprise partnerships, the competitive advantage shifts permanently toward established players over independent innovators

Opposed

The 'too dangerous to release' framing is primarily marketing, reminiscent of OpenAI's similar claims about GPT-2 that proved overblown — Mythos will likely be released once the hype cycle is complete
Anthropic probably lacks the compute to serve Mythos at scale, making the safety narrative a convenient cover for economic limitations rather than a deliberate neofeudal strategy
Open-source models with proper orchestration can already replicate much of what Mythos achieves, making the gap less consequential than the article suggests
The restriction is better understood as responsible disclosure — giving security teams time to patch vulnerabilities before a powerful exploit-finding tool becomes public
Existing models remain highly capable and available; nothing has been taken away from individual developers who can still build freely with current tools
Competition between AI labs and the open-source ecosystem will naturally push capabilities downward to the public within months, not create a permanent access divide