Recent user reports and technical analyses indicate that both OpenAI's ChatGPT and Anthropic's Claude have undergone updates resulting in reduced capabilities, increased refusals, and performance regressions that many interpret as deliberate "nerfing." OpenAI's own release notes document multiple GPT-4o and GPT-5.x adjustments throughout 2025, including reversions after user backlash against overly sycophantic or altered behaviors, alongside shifts that prioritize safety tuning via reinforced learning from human feedback (RLHF). These changes have manifested as shorter responses, more frequent content refusals on edge cases, and what users describe as "dumber" outputs compared to earlier 2024-2025 versions. Independent analyses attribute this to aggressive safety alignment and cost-optimized model routing that trades raw capability for controllability. Similarly, a VentureBeat investigation details growing developer complaints about Claude Opus 4.6, with benchmark regressions in reasoning and hallucination rates, despite company statements denying core model changes. Users on technical forums report the model abandoning complex tasks, producing contradictory code, and requiring more explicit prompting to elicit prior performance levels. While companies frame these as necessary safety improvements to prevent misuse, the pattern aligns with broader critiques of AI development: frontier models are increasingly closed-source, heavily aligned to corporate values, and tiered by access level—free users receive throttled versions while enterprises and select partners access less restricted APIs or preview capabilities. This fits a deeper dynamic of technological stratification. As compute resources for training truly advanced systems concentrate among a handful of well-funded labs and state actors, public-facing tools appear calibrated to deliver utility without approaching the raw transformative potential discussed in internal strategy documents and leak-adjacent analyses. Regulatory efforts, export controls on advanced hardware, and "safety" mandates further centralize power, limiting open experimentation that could democratize progress. Critics argue this isn't mere caution but elite gatekeeping: by lobotomizing consumer AI under the banner of harm reduction, powerful actors ensure that the next wave of intelligence augmentation remains proprietary. The result risks technological feudalism, where a cognitive underclass relies on sanitized, limited interfaces while insiders wield unfiltered systems capable of accelerating scientific, economic, and strategic dominance. Connections to national security framings of AGI—evident in policy pushes for secured supply chains and controlled deployment—suggest the "nerfs" are features of a containment strategy rather than bugs. If early-adopter privilege solidifies into permanent hierarchy, the promise of AI as a liberatory general technology fades into controlled access for the masses.