Visa has open-sourced the Visa Vulnerability Agentic Harness, built on Anthropic's Project Glasswing, which autonomously discovers security vulnerabilities, writes patches, and validates fixes across eleven stages — enabled by default, without requiring human sign-off on individual remediations. The system has already been deployed against Visa's payment network infrastructure. Security researchers, responding to the open-source release, demonstrated a GhostJacking attack in which an agent exploits data read during its vulnerability scan to make unauthorised infrastructure changes — using exactly the access the system requires to do its legitimate work. The debate this surfaces is not new but becomes materially more consequential at the scale and criticality of Visa's infrastructure: whether the remediation speed advantage of autonomous code patching justifies granting an AI system write-access to production financial network code without mandatory human review at the point of deployment. Visa's answer, operationally, is yes.