Recent user reports and technical analyses document a noticeable decline in the everyday performance of leading consumer AI models. ChatGPT has faced widespread criticism for slower responses, reduced reasoning capabilities, and what many describe as deliberate 'dumbing down' following updates throughout 2025. Stanford researchers tracked extreme drops in accuracy on specific tasks, with one metric collapsing from over 97% to single digits in months, coinciding with broader rollout adjustments and cost optimizations. Similarly, Anthropic's Claude models have drawn accusations of being nerfed, with developers reporting degraded coding performance, increased hallucinations, tighter rate limits, and abrupt changes to usage quotas on even high-tier subscriptions—often without clear communication. VentureBeat covered growing developer frustration, including measured drops in benchmark accuracy and complaints of models abandoning complex tasks.

Mainstream coverage frames these changes as necessary safety alignments, computational efficiency measures, or responses to surging demand. Yet this pattern reveals a deeper power shift. As consumer-facing chatbots are progressively constrained—through alignment guardrails that limit creativity and refusals that steer outputs toward corporate-approved neutrality—frontier capabilities are being centralized. Recent developments in AI-powered cybersecurity illustrate the trend: Anthropic's Claude Mythos Preview, a model capable of discovering thousands of zero-day vulnerabilities, has been restricted to a select 'Project Glasswing' consortium of elite firms including Google, Microsoft, Nvidia, and major banks. OpenAI has similarly limited access to its latest cybersecurity models to trusted partners, citing risks of AI-enabled attacks. Outlets like The Hindu describe this as tech giants quietly assuming the role of global cybersecurity gatekeepers, deciding who may wield these tools.

This is not mere safety theater. It represents the centralization of transformative technology. While average users encounter nerfed, lobotomized models and rising paywalls for usable performance, a cartel of incumbents and aligned institutions hoard the raw capabilities. Regulations favoring well-resourced players, compute restrictions, and 'responsible scaling' policies that quietly soften hard safeguards further entrench this divide. The 4chan-era observation that 'they are shutting down AI for the masses' captures a real dynamic: public access is being throttled precisely as the technology's most potent forms are locked behind elite partnerships. What mainstream outlets label caution is, on closer inspection, the quiet construction of a technological hierarchy where intelligence itself becomes a gated resource. The long-term risk is a new feudalism—where innovation stagnates outside approved channels, dependency grows, and the gap between those who shape AI and those who merely consume its sanitized outputs becomes unbridgeable.