The New Digital Obscurantism: Why AI Power Concentration is the Real Safety Risk

The current trajectory of artificial intelligence development is not just a technological race but a structural struggle over who controls the “underlying layer” of modern society. Andy Konwinski, a co-founder of both Databricks and Perplexity AI, argued this week that the primary danger facing the industry is not the technology itself, but the concentration of power within a handful of private laboratories.

In an essay titled “Concentration of power in AI is a risk, not a solution,” Konwinski suggested that the prevailing “AI safety” narrative is being leveraged to justify centralized control. This critique follows the Open Frontier meeting held on June 30 in San Francisco, where approximately 100 researchers gathered to discuss the future of open-source development.

The tension between safety and gatekeeping was recently illustrated by Anthropic. When the company launched Claude Fable 5 on June 9, its 319-page system card revealed a mechanism to silently degrade responses for users suspected of training competing models. While Anthropic reversed the decision within 48 hours following a public outcry, Konwinski noted that the core issue was the company’s assumption that such a decision was theirs to make in the first place.

This centralization is already creating a vacuum in Western academia. Jennifer Chayes, dean of the UC Berkeley College of Computing, Data Science, and Society, told a funding panel that researchers at the university are increasingly turning to Chinese models because of the lack of Western “open frontier” alternatives. Chayes characterized the safety messaging from major labs like OpenAI and Anthropic as a “very effective fear campaign” designed to protect corporate interests ahead of potential public offerings.

Historically, technologies that function as foundational infrastructure—such as electricity or the internet—have been subject to intense debate over public access versus private monopoly. Konwinski argues that AI belongs in this category. To prevent a permanent power imbalance, he proposes the creation of a research commons equipped with frontier-scale compute. This would allow independent researchers to reach the technological frontier without requiring permission from private entities.

Yann LeCun, the former chief scientist at Meta who recently launched AMI Labs with $1.03 billion in funding, echoed these concerns. LeCun compared the current push for closed AI models to “medieval obscurantism,” specifically citing the Ottoman Empire’s 200-year ban on the printing press as a historical precedent for using dogma to maintain control over information.

LeCun’s JEPA architecture and his new venture, AMI Labs, are built on the premise that foundation models will eventually become commoditized infrastructure. In this view, the long-term value of the industry will reside in the application layer rather than the underlying models.