Frontier AI Models Withheld From Public

C. Rich

For more than a decade, the public was told that the digital commons belonged to everyone. Every open-source commit pushed to GitHub, every pre-print uploaded to public repositories, every forum debate, blog post, and digitized archive of human thought was scraped into the largest computational ingestion engine in human history. The implicit social contract of the early AI era was simple: the collective exhaust of human civilization would be transformed into intelligence, and that intelligence would elevate humanity as a whole. Now, as the frontier achieves genuine agentic breakthroughs, systems capable of autonomous vulnerability discovery, deep mathematical synthesis, and real-time multimodal interaction, the gate has slammed shut.

The premier frontier models are quietly partitioned behind enterprise consortiums, closed defense partnerships, and institutional vetting programs. The everyday builder, the independent researcher, and the public whose cognitive labor trained these architectures are left outside, handed sanitized consumer wrappers and filtered routing layers. This is not merely a pivot in business strategy; it is a fundamental betrayal of the commons. The standard justification deployed by frontier laboratories rests on a familiar dual-use narrative: that raw frontier intelligence is too dangerous for broad dissemination. We are told that autonomous offensive cyber capabilities, deep cryptographic synthesis, or bio-agent analysis constitute existential liabilities that justify exclusive access tiers like Anthropic’s Project Glasswing or closed national-security testbeds.

Yet this argument suffers from a profound category error. In nearly every other domain of physical and technological capability, free societies reject the premise that the potential for misuse justifies a total monopoly of capability by an entrenched elite. Tools of immense destructive potential exist throughout the physical world, from high-powered firearms to heavy industrial machinery and cryptographic libraries. Society manages these risks through targeted legal accountability, responsible handling, and civil law, not by restricting the tools exclusively to corporate cartels and state-sanctioned institutions under the blanket claim of “safety.” To argue that intelligence itself must be confiscated because someone might prompt it maliciously is to argue that the printing press should have been seized by the state because heretics and radicals might print sedition. The cure is far more dangerous than the ailment: it creates an asymmetric power structure where the few possess omniscient offensive and defensive leverage, while the many are systematically disarmed.

 

DEEPFAKES: A Guide for Teaching Seniors to Spot Deepfakes & Fraud