Tuesday was a dark day. With the release of Fable 5, we saw another step in a trajectory that should worry anyone who cares about open science, individual agency and the future distribution of power. Fable appears to restrict or silently sabotage many ordinary requests related to AI research, security research, biology research and other domains of scientific inquiry.

The AI revolution was built on the data scraped from internet commons, open scientific publishing in peer-reviewed conferences, open-source frameworks such as PyTorch, public benchmarks starting with ImageNet. But once these closed labs have reached the Recursive Self-Improvement (RSI), the ladder is being pulled up behind them.

If you are outside Anthropic (and soon maybe OpenAI or any other closed frontier lab, they have a precedence to do the same), you are increasingly blocked from using the most advanced systems for AI research. The smokescreen of justification is usually safety, geopolitical competition, biosecurity and cybersecurity. But the effect is the same: concentration of power to a small number of companies.

If we apply a scaling-law mindset to this institutional pattern, the obvious question is: where does the ladder disappear next? Closed labs are already moving toward mathematics, biology, materials science, and the natural sciences more broadly (see their emphasis on AI-for-Science). As these labs build RSI pipelines, they will increasingly improve the machinery of discovery in these domains itself. If access to the most capable systems in these fields is controlled by a small number of closed labs, then we should expect a future where the scientific agency itself becomes centralized. Unless you are INSIDE one of these organizations, FUNDED by them, or operating under their PERMISSION structure, you may not be able to pursue frontier research at all.

That is the deeper danger: THE LADDER IS GOING AWAY.

Earlier technologies had slower development cycles, engineers moved between companies which led to diffusion of trade secrets and spread of tacit knowledge. Universities, startups, and independent researchers could still catch up. But AI changes the tempo. The result is not merely “closed AI.” It is potentially CLOSED SCIENCE.

I want to put this bluntly. Dario Amodei and others at Anthropic have repeatedly presented themselves as the only good guys on this planet who can be stewards of AI. This position comes off as filled with HUBRIS. Who the fuck are they to dictate about my freedom or for rest of the world? Just building the most advanced AI system currently doesn’t give them the license to dictate freedom for me, my family, my descendants or for anyone for that matter. A world of EXTREME POWER CONCENTRATION is coming soon, unless an effective counter is mounted.

And this will not be unique to Anthropic. Every closed frontier lab has the precedence now to converge on the same posture: use the open commons to climb, restrict access at the frontier, justify closure through safety and accumulate the economic and scientific returns internally.

AI safety discourse has often focused on existential risk from superintelligence. People like Yudkowsky, Soares, Scott Alexander, and others have spent years emphasizing the possibility of human extinction. I do think that concern has merits but the more immediate and legible risk is hyper-centralization: a narrow group of institutions accumulating unprecedented wealth, scientific capability, political leverage, and control over the tools of thought itself. That second scenario deserves far more attention than it has received from these countless years of debate about AI safety.

I was personally sad when Andrej Karpathy joined Anthropic. For years, Andrej represented something I deeply admired: the spirit of explaining, teaching, and making modern AI legible to the broader technical world. Seeing him join one of the most closed and open-source-hostile frontier labs felt symbolically painful.

Of course, individuals have to make their own choices. It is rational to look after yourself and your family. If the world is moving toward concentrated power, many people will try to join the winning side. I do not blame anyone for feeling that pressure. But I also believe the impact of a life should be larger than merely securing a place inside the dominant institution.

At moments like this, I think it is important for those of us at Eigen Labs to return to the mission of the company:

We are a research lab building coordination tools that preserve and expand individual agency in a post-AGI world. We pursue long-horizon research across three domains: we investigate the technologies and incentive systems that enable human and agent coordination; we prototype tools that can empower individuals and open networks to coordinate and compete at the speed of machine intelligence; and we build the new verifiable infrastructure needed to govern and constrain agent action in service of human progress.

This is not a corporate word salad. This mission matters precisely because the default trajectory of AI is centralization. AI is one of the most profound technologies humanity has ever built, comparable to fire, the wheel, the printing press, and flight. But if this technology is not used to expand human agency, then what exactly are we building? If AI becomes a machine for enclosing knowledge, concentrating power, and making billions of ordinary people dependent on a few institutions, then there is something wrong with this world.

Eigen Labs has to exist to push in the opposite direction. We have already taken early steps. Darkbloom showed that individuals with ordinary hardware can participate in inference supply. ecdsa.fail points in another direction: open agentic science where domain experts, citizen scientists and others came together to collaborate using their agents to beat Google’s CENSORED result.

These may be early products, but they are not small ideas. They are examples of a broader philosophy: build systems where individuals have agency.

That is the work ahead. That is why Eigen Labs’ mission matters. THE LADDER IS GOING AWAY. OUR JOB IS TO BUILD NEW LADDERS.