Cisco and IBM plan distinct cybersecurity uses for Anthropic’s Claude Mythos, extending the model into AI testing and customer code scans as businesses add agents to core workflows.
IBM’s Concert Protect would scan code repositories chosen by customers and assess whether weaknesses could be exploited. The system would make proposed edits available, while developers retain the choice to accept or reject them. IBM has used Mythos for internal security research and evaluation; the planned integration would bring it into enterprise customer workflows.
Cisco’s AI Defense would use Mythos to develop multistep attack exercises tailored to the AI application or agent being examined. Customers would define each engagement’s target and objective; activity would stay within their authorization and be attributable to a named operator. Sessions would be logged, and customers could stop a test or let it expire. The plans build on Cisco’s earlier AI Defense work.
Anthropic Cybersecurity Lead Michael Moore said companies want to check AI agents before putting them into production as the tools take on core tasks. Cisco joined the organizations selected for the first Glasswing work; Anthropic announced the defensive-security initiative in April. Anthropic describes Mythos’s biology and cybersecurity capabilities as dual-use; access is limited to organizations it has screened.
The announcements include no measurements of detection or remediation performance and no customer case studies. Cisco expects to begin with selected enterprise customers soon, while IBM is targeting general availability in early 2027.
