Visa’s technology team quickly grasped the security challenges posed by advanced AI models when they began testing Anthropic’s Mythos model. During the early stages of Project Glasswing, the team noticed how swiftly attackers could pinpoint and exploit vulnerabilities in critical code bases, which heightened security risks, as explained by Rajat Taneja, Visa’s president of technology, during a preparation call for his upcoming session at VB Transform 2026, VentureBeat’s agentic AI event.
Visa was selected among other companies to test Anthropic’s new model. However, a version of this model, released on June 9, was soon disabled to adhere to U.S. government directives.
The insights from Project Glasswing highlighted the growing gaps in enterprise security and the opportunities these present for malicious actors. Taneja noted to VentureBeat that, “Security has always been important, but currently, in the age of AI, it is going to be even more important because the attacks become autonomous. The defenses have to become autonomous. And we are not there. And there’s an asymmetry there, which is a very big risk for the world.”
According to Cisco’s State of AI Security 2026 report, threat actors now possess powerful AI agents capable of operating non-stop, “operating at a scale and speed that human teams cannot match, automating the tedious reconnaissance and exploitation phases of a cyberattack.” Amy Chang, Cisco’s head of AI threat intelligence and security research, is also scheduled to speak at VB Transform.
To counter these risks, Visa is developing its own abstraction layers, observability tools, and data guardrails to protect its autonomous commerce frameworks. The company has also launched an open-source, AI-driven security framework that transforms vulnerability detection and remediation into a systematic, repeatable process.
This initiative reflects a necessary shift for enterprise IT teams to safeguard systems against threats from malicious actors using autonomous agents. Taneja will present these insights and technical details in his session at VB Transform, titled Inside Project Glasswing and Mythos: Securing the agentic future today, scheduled for July 15.
Additional sessions at VB Transform focusing on agentic AI security include:
-
CrabTrap: How Brex built an open source proxy to secure OpenClaw’s critical flaws for everyone with Brex co-founder and CEO Pedro Franceschi;
-
When AI Agents have wallets: Building the trust layer for autonomous B2B commerce with Mastercard’s Chief AI and Data Officer, Greg Ulrich;
-
Expedia’s blueprint for building autonomous agents for high-stakes transactional systems with Chief AI and Data Officer Xavier Amatrain; and
-
Securing agentic AI: A playbook for permissioning, sandboxing, and human-in-the-loop controls, a panel discussion with AI security leaders from Intuit, Box and Cisco.
Interested in attending VB Transform 2026? Register here. A select number of complimentary passes are also available to senior technology leaders. Contact us to get yours.
