Replymessage unavailable
Interestingly enough in Anthropic’s April 7, 2026 Mythos Preview cybersecurity report, they disclose that the impressive vulnerability-finding results were produced with very large-scale repeated runs, not a single cheap Claude session. For example, Anthropic says it ran its OpenBSD vulnerability-search scaffold 1,000 times. Those runs cost just under $20,000 in API-equivalent usage and produced several dozen findings.