Visa opens AI cyber defences, lets agents make payments

Visa open-sourced part of its AI-powered cyber defence system this week after the company's president of technology acknowledged the firm had been humbled by vulnerabilities exposed through Anthropic's Mythos model and an AI-agent attack on Hugging Face. In the same announcement, Visa confirmed it has begun allowing certain AI agents to transact independently using Visa infrastructure, a quiet but consequential shift in how money can move. Regulators at GFF 2026 pushed back the same week, with the NPCI, SEBI and a Swiss bank's AI chief all agreeing that agents should determine intent but a separate authorisation layer must control whether funds actually move.