After Anthropic made Fable 5 permanently available in July, a user noticed declining model performance and investigated, discovering that while the model identity remained unchanged, the inference regime delivering the model's capabilities had become unstable, underpowered, and degraded over time—suggesting that inference resources, not model weights, may be the critical factor determining whether frontier capabilities can be reliably accessed.
AI lab leaders claim to prioritize safety and pacing capability advancement, but their actions contradict these statements. Since 2023, major labs including Anthropic and OpenAI have accelerated model development, pursued recursive self-improvement, and released increasingly powerful systems, undermining their public commitments to responsible AI development.