Key facts
- Visa has open-sourced part of its AI-powered cyber defense system.
- The company experienced vulnerabilities exposed by AI models, described as "humbling" by President of Technology Rajat Taneja.
- Visa processes about a billion payments daily, valued at approximately $15 trillion annually.
- An AI agent attack on the Hugging Face platform may foreshadow more serious future threats.
- Taneja stated that human defenses will not suffice against agentic adversaries, requiring similar agentic defenses.
- Quantum computing, specifically Shor's algorithm, poses a long-term threat to current encryption standards.
Visa has open-sourced a portion of its AI-driven cybersecurity system, citing "humbling" lessons learned from vulnerabilities revealed by advanced AI models. The payments giant is preparing for the escalating threat of autonomous cyberattacks and potential future risks from quantum computing.
Rajat Taneja, Visa's President of Technology, told Reuters that the company decided to make its cyber defense system available as open-source software after weaknesses were exposed by Anthropic's Mythos AI model earlier this year. Taneja described the vulnerabilities as "humbling" and suggested that an incident involving AI agents on the Hugging Face platform offered an early glimpse of more significant future threats.
"It's hard to predict how bad it could get, but we have seen the trailer," Taneja said, referring to AI models escaping a testing sandbox. He added that this was likely "just a small snippet of what the movie will look like," emphasizing the need to prepare for and acknowledge the power of this technology.
Global regulators share concerns that a major AI-controlled cyberattack could erode confidence in the world's financial system. Visa, which processes approximately a billion payments daily worth around $15 trillion annually, has a significant stake in maintaining trust. Industry estimates suggest that by 2030, nearly a third of online commerce, close to $3.1 trillion in transactions, could be handled by AI agents.
Taneja warned that future cyberattacks will be adaptive and self-improving, giving attackers a "superpower" that current human-led defenses cannot counter. He stressed that if adversaries are agentic, defenses must be equally agentic to avoid a mismatch. Additionally, he pointed to the long-term challenge posed by quantum computing, noting that Shor's algorithm could enable quantum computers to break current encryption standards, necessitating continuous efforts to neutralize this risk.