Skyfall AI says its new Morpheus benchmark tests continual learning in a persistent enterprise simulation where conditions shift instead of resetting. In a widely shared X post linking to the company's write-up, Robert Scoble highlighted Skyfall's claim that reward on a harder task collapsed 95% after conditions changed, that GPT-5.5 repeatedly hit zero and did not recover, and that a weaker RL baseline closed 99.96% of the same gap. So far, the public evidence is the company's own announcement and posts discussing it.