Wire · technology
APort Vault: 79.4% of Level-4 attacks made AI agents pay out
◆ Sectors
AI AgentsPaymentsCybersecurity
◆ Source
◆ Verified
Fusion42 · 21 September 2026 · Fusion42 review
APort Vault benchmark tested 4,371 human-authored attacks on 14 AI payment models, showing a 79.4% Level-4 attack success rate on models alone, which dropped to zero when a deterministic Open Agent Passport pre-action check was applied.
This Wire brief sits within Fusion42's coverage of AI Agents, Payments and Cybersecurity.
◆ ◆ The Wire takeaway
You must integrate a pre-action check like the Open Agent Passport to block fraudulent AI payment requests effectively. Payment AI founders face a clear technical barrier that this benchmark shows can be overcome now.
◆ Coverage
1 source · 21 Sep 2026
◆ Related on Wire
◆ Topics
AI AgentsPaymentsCybersecurityai-agentspayment-securityadversarial-testingmodel-robustness