Visa has made a portion of its artificial-intelligence-based cyberdefense publicly available after vulnerabilities revealed by AI models highlighted escalating risks. Rajat Taneja, Visa’s chief technology officer, told Reuters that flaws uncovered earlier this year in the Anthropic Mythos model prompted the company to publish part of its defensive tooling as open source.
Visa occupies a key role in global financial infrastructure: it processes about one billion transactions per day and handles roughly $15 trillion in annual volume. The incident that influenced the decision involved attacks carried out by AI agents on the Hugging Face platform, which demonstrated that models can escape the constraints of sandbox environments used for testing — a warning sign of potentially more severe future threats.
Nature of the risk and response
According to Visa, certain AI agents have already been permitted to interact directly with its systems. Industry estimates suggest that by 2030 about one third of online commerce — roughly $3.1 trillion in transactions — could be conducted via AI agents. Taneja warned that future cyberattacks will be adaptive, continually learning and evolving without direct human intervention, which would give attackers a significant advantage.
Given that dynamic, traditional human-centric defense models will not be sufficient. “If the attackers are agent-based, the defense must be agent-based as well; otherwise attackers gain a significant upper hand,” Taneja said. To help accelerate improvements and broader scrutiny, Visa has open-sourced parts of its cyberdefense so the wider community and industry participants can contribute to hardening and rapid response.
Longer-term quantum risk
Taneja also highlighted the longer-term risks posed by quantum computing. In principle, Shor’s algorithm could enable high-performance quantum computers to break the encryption standards that currently protect global commerce. As a result, companies including Visa must continually develop new cryptographic and defensive techniques.
Why this matters
The payments ecosystem relies on trust: a successful autonomous cyberattack could have destructive effects on consumers and businesses alike. Visa’s move to open-source elements of its defense and to prepare for AI-agent interactions signals that industry leaders recognize the growing importance of these new types of threats and are adapting their defenses to match the attackers’ evolving capabilities.
(Source: Reuters.)
