Anthropic Mythos's showcase vulnerabilities were tested on small, inexpensive, open-weight models, revealing similar analysis results. AI cybersecurity capability varies significantly with model size, indicating that the security moat relies on the system architecture rather than the model itself.