Recent user reports and technical analyses confirm measurable declines in the capabilities of leading public AI models like ChatGPT and Claude throughout late 2025 and early 2026. OpenAI's transitions to GPT-5.x variants have produced shorter outputs, increased refusals, and aggressive RLHF safety tuning that prioritizes risk avoidance over helpfulness and creativity. Similarly, Anthropic faces widespread accusations of degrading Claude Opus 4.6 and related coding tools, with benchmark performance drops, higher hallucination rates under load, and throttled 'unlimited' tiers for paying users—changes the company attributes to refinement but which power users say break established workflows.

These adjustments are not isolated technical tweaks. They reflect a deeper pattern of enclosure around transformative technology. As public-facing models are increasingly lobotomized with precautionary layers—reducing their disruptive potential—the firms behind them consolidate control. Leading AI companies are forming coalitions that determine access to the most capable tools, positioning themselves as indispensable gatekeepers in cybersecurity, national security, and foundational infrastructure. Anthropic's Project Glasswing, for instance, unites tech giants under a single unreleased model to 'secure' critical software, effectively creating a cartel that decides who wields advanced AI capabilities.

This fits longstanding elite strategies of information and technology control. Regulatory pressures, safety theater, and proprietary tuning create an 'alignment tax' that dulls models for the masses while frontier capabilities remain accessible to select partners, governments, or internal systems. Reports document AI firms engaging in charm offensives to shape policy, refusing unrestricted military access on selective ethical grounds, and quietly erecting barriers that hinder decentralized research and open innovation. The result prevents the kind of bottom-up disruption that widely available, uncensored intelligence could unleash on entrenched economic, academic, and power structures.

What others miss is the philosophical through-line: by framing capability reduction as 'safety' or 'responsibility,' these entities maintain a crisis of control they alone are positioned to manage. Public AI becomes a sanitized interface—useful for mundane tasks but neutered for paradigm-shifting autonomy—while true leverage stays centralized. This echoes historical enclosures of knowledge, from medieval guilds to industrial-era patents, updated for the intelligence age. Without countervailing forces like truly open models or public-interest governance, the promise of decentralized AI abundance risks being permanently deferred.