The safety implications of superior AI fashions had been instantly clear to Visa’s expertise group after they started testing Anthropic’s Mythos mannequin.
Simply weeks into Venture Glasswing, the group noticed how shortly attackers can establish and weaponize vulnerabilities in important code bases, creating safety dangers, defined Rajat Taneja, Visa’s president of expertise, throughout a name to organize for his session at VB Rework 2026, VentureBeat’s upcoming agentic AI occasion.
Visa is among the many corporations chosen to check Anthropic’s upcoming mannequin — a model of which was launched June 9 however abruptly disabled days later to adjust to U.S. authorities directives.
The findings of Venture Glasswing put a highlight on widening enterprise safety gaps and the vulnerabilities malicious actors can make the most of.
“Safety has at all times been necessary, however at the moment, within the age of AI, goes to be much more necessary as a result of the assaults turn into autonomous,” Taneja informed VentureBeat. “The defenses should turn into autonomous. And we aren’t there. And there is an asymmetry there, which is a really massive danger for the world.”
Risk actors now have entry to highly effective AI brokers that may work 24/7, “working at a scale and pace that human groups can’t match, automating the tedious reconnaissance and exploitation phases of a cyberattack,” in accordance with Cisco’s State of AI Safety 2026 report. Amy Chang, Cisco’s head of AI risk intelligence and safety analysis, will even be a speaker at VB Rework.
To mitigate these dangers, Visa is constructing its personal abstraction layers, observability, and knowledge guardrails to safe its autonomous commerce frameworks. The cost providers big additionally rolled out an open‑supply, AI-driven safety framework that turns vulnerability discovery and remediation right into a structured, repeatable pipeline.
Their work represents a shift enterprise IT groups should make to guard enterprise methods in opposition to threats posed by dangerous actors wielding autonomous brokers. Taneja will share these insights and precious technical particulars throughout his session at VB Rework, titled Inside Venture Glasswing and Mythos: Securing the agentic future at this time, on July 15.
Different agentic AI security-focused classes at VB Rework embody:
-
CrabTrap: How Brex constructed an open supply proxy to safe OpenClaw’s important flaws for everybody with Brex co-founder and CEO Pedro Franceschi;
-
When AI Brokers have wallets: Constructing the belief layer for autonomous B2B commerce with Mastercard’s Chief AI and Information Officer, Greg Ulrich;
-
Expedia’s blueprint for constructing autonomous brokers for high-stakes transactional methods with Chief AI and Information Officer Xavier Amatrain; and
-
Securing agentic AI: A playbook for permissioning, sandboxing, and human-in-the-loop controls, a panel dialogue with AI safety leaders from Intuit, Field and Cisco.
Curious about attending VB Rework 2026? Register right here. A choose variety of complimentary passes are additionally obtainable to senior expertise leaders. Contact us to get yours.