Wire · operational-macro
AI Agents Fail Nearly Half of Shopping Tasks in Visa Test
◆ Sectors
◆ Geography
◆ Source
◆ Verified
Fusion42 · 5 November 2026 · Fusion42 review
Visa's new Agentic Commerce Benchmark tested 22 AI models on shopping tasks, revealing a 57.2% average success rate with notable variability in performance and instability. Trust in AI for payments remains low among consumers, prompting Visa and partners to develop verification rules and tools to assign responsibility and improve security for AI agent commerce.
This Wire brief sits within Fusion42's coverage of Fintech.
◆ ◆ The Wire takeaway
Visa's push for AI agent verification and security standards sets new compliance bars for fintech startups leveraging AI for commerce. You need to prepare for clearer rules on responsibility and consumer trust hurdles as AI agents handle payments.
◆ Coverage
1 source · 5 Nov 2026
◆ Related on Wire
◆ Topics