AI security startup Depthfirst says its specialized model discovered critical vulnerabilities missed by Anthropic's Mythos at a fraction of the cost, including a flaw lurking in NGINX for 18 years, and launches a $5M open defense initiative.