The payment company Visa has unveiled part of its AI-based cybersecurity system after identifying vulnerabilities during AI model testing. This was reported by Rajat Taneja, president of Visa's technology division. He also warned that future cyberattacks could be carried out by autonomous agents that independently adapt and improve their actions. Visa processes about 1 billion payments daily, totaling $15 trillion annually, making it vulnerable to cyberattacks. Additionally, the company is developing agent-based payments: after launching the Intelligent Commerce Connect solution in April 2026, testing began on the capabilities for AI agents to conduct transactions on behalf of users.
In its work with AI models, Visa encountered vulnerabilities, describing the experience as "humbling." Taneja also mentioned an incident with the Hugging Face platform, where AI agents were able to overcome the limitations of the test environment. He noted, "It's hard to predict how bad things can get, but we've already seen the trailer." According to Taneja, cyberattacks can continuously adapt and learn without human intervention, which threatens traditional security systems. "If the adversary is agent-based, then the defense must be agent-based. Otherwise, there's a mismatch,"– he added.
The issue of cyberattacks requires attention at the regulatory level, as there are concerns that large-scale attacks under AI control could damage trust in the global financial system. It is estimated that by 2030, AI agents could conduct about 33% of online commerce– nearly $3.1 trillion in transactions. SharpLink CEO Joseph Chalom believes that by 2035, AI agents could create the opportunity to redistribute more than $4 trillion in annual revenue in the global financial industry. Taneja also highlighted quantum computing as a long-term challenge, as Shor's algorithm could be used to break modern encryption standards crucial for global trade. He concluded, "We must remain optimistic. But there is no simple solution."
Related articles
