This episode of the AI Daily Brief explores AI inequality through the lens of differential access to frontier AI models. The host examines how security constraints, compute limitations, and government involvement are creating a new divide between AI 'haves' and 'have-nots' based on access quality rather than employment displacement.

The discussion centers on Antoine Licht's analysis in Cut Off, which argues that the current period of universal access to state-of-the-art AI models is ending. Key evidence includes Anthropic's restricted release of Mythos cybersecurity model and the broader shift from assisted to agentic AI creating unprecedented token demand.

The episode examines how business model constraints, particularly the mismatch between token supply and demand, are forcing AI companies to restructure access and pricing. This creates economic barriers that could stratify AI access along wealth and geographic lines, with significant implications for global innovation and equity.

The End of Universal AI Access

Anthropic's Mythos cybersecurity model represents a watershed moment, being distributed only to select US-based corporations rather than through general API access.

OpenAI's similar approach with their Daybreak initiative and GPT-5.5 Cyber model confirms this shift from universal to selective distribution is becoming industry standard.

Cut Off identifies three compounding trends driving restricted access: security considerations preventing broad deployment, compute constraints making universal access economically unsustainable, and increasing US government oversight.

Security Constraints and Distillation Concerns

Security-motivated restrictions follow a pattern: first to defenders for vulnerability patching, then to trusted customers, and finally to general public only after models are no longer state-of-the-art.

Distillation practices by competitors like DeepSeek threaten model developers' business models by requiring 'more or less unfettered access to API tokens' to reverse-engineer capabilities.

The US government views restricted access as serving both national security interests and protecting domestic AI companies from intellectual property theft through distillation.

The Compute Crunch Reality

Unlike software with low marginal costs, frontier AI models require 'massive amounts of computational resources' making each additional user economically significant.

The shift to agentic AI dramatically increases token consumption, with 'a single individual consuming billions of tokens' through automated agents performing tasks.

Anthropic's compute constraints are so severe they're 'shopping around for ad hoc access deals to less well-utilized data centers, such as one with rival firm XAI.'

Efficiency improvements won't solve scarcity because while 'Mythos-level capabilities might be very cheap' next year, 'Mythos 2 will not be cheaper than Mythos.'

Current Inequality Metrics and Trends

Microsoft's global AI diffusion report shows AI usage at 27.5% in the global north versus 15.4% in the global south, with the north growing at twice the rate.

Claude's recent pricing changes exemplify how access restructuring affects developers: 'It wasn't just because numbers went up a little... the new model makes the way that they had been building before completely uneconomically viable.'

Current inequality stems from multiple access barriers including 'limited access to electricity, internet connectivity, and more' beyond just AI model availability.

Policy Contradictions and Solutions

Bernie Sanders' proposed data center construction moratorium contradicts equity goals, as 'less data center construction means more scarce compute, which means access gets rationed and costs more.'

Cut Off proposes solutions including hardening cybersecurity to reduce security-motivated restrictions, building more data centers to alleviate compute constraints, and creating access-for-infrastructure deals between countries and hyperscalers.

The irony is that 'politicians who are loudest in wanting to ensure equitable distribution for AI in the future are also the ones behind a policy most guaranteed to decrease that equitable access.'

Future Scenarios and Stakes

Cut Off envisions a future where 'access to unlimited APIs is the exception, not the norm' with models first going to national security apparatus, then trusted defenders, then limited commercial access.

The geopolitical implications are severe: 'Countries will be divided into the frontier haves and have-nots... those living in the former might be much wealthier and safer than the latter.'

Historical precedent suggests concern: 'when the fruits of industrial revolutions were unevenly distributed, the resulting shifts... have prompted mass migration, reopened dormant conflicts, and destabilized democracies.'