Palo Alto Networks warns that automated interactions may create hidden backdoors, requiring enterprises to implement governance-oriented browsers to establish monitoring mechanisms, thus comprehensively testing their governance capabilities in the AI-native era.

As generative AI rapidly gains popularity, enterprises are moving from "AI-assisted" to "AI-native" technologies. This is because, with the rise of agentic AI, enterprise work patterns have shifted from "human-to-application" to automated "agent-to-agent" interactive networks. These machine-speed-operating behaviors are gradually moving beyond the scope of traditional cybersecurity monitoring, becoming emerging sources of risk. However, how should enterprises prepare for this new cybersecurity challenge?

Palo Alto Networks states that the risks of agent-based AI primarily stem from two scenarios. The first is "human-to-agent," where employees input sensitive data into large language models for analysis, potentially leading to the leakage of corporate secrets. The second, more difficult-to-detect, is "agent-to-agent" interaction, where, without human oversight, different system agents may automatically collaborate and grant excessive privileges, creating potential cybersecurity vulnerabilities. Because each agent possesses an identity and permissions, this behavior is equivalent to establishing an uncontrolled "invisible backdoor" within the enterprise.

Under this trend, the role of browsers has also undergone a fundamental transformation. Hsiao Sung-ying, Senior Technical Director at Palo Alto Networks Taiwan, points out that while enterprises previously paid less attention to this aspect, Palo Alto Networks observes that browsers have evolved from simple network gateways into core interfaces integrating workflows and proxy operations, and are even considered operating platforms for modern work. If enterprises do not simultaneously implement governance and monitoring mechanisms, proxy browsers may amplify risks, creating another cross-organizational security gap. Therefore, establishing a "governance-oriented browser" with consistent control capabilities has become crucial for enterprises to manage proxy behavior.

In light of this, and in response to the challenges posed by agent-based AI, Palo Alto Networks proposes a "Governed Autonomy" architecture, emphasizing three core elements: traceable identity, clearly defined operational boundaries, and contextual judgment capabilities. By bringing agent behavior into an auditable and controllable framework, enterprises can improve efficiency while avoiding the risk of losing control.

To help enterprises address the various new cybersecurity issues arising from agent-based AI and to further effectively manage and supervise agent behavior, ensuring that enterprises can balance innovation and security during digital transformation, Palo Alto Networks has launched corresponding products. Hsiao Sung-ying explained that for enterprises transitioning to "governed autonomy," Prisma AIRS 3.0 can help enterprises expand from observing AI behavior to controlled and governed autonomous execution by protecting the entire lifecycle of agent-based AI.

Regarding browser security, Palo Alto Networks has launched an upgraded Prisma Browser specifically designed for agent-based AI. Hsiao Sung-ying explained that Prisma Browser transforms the web environment into a secure, AI-driven work platform, allowing users to significantly improve productivity through agent-based AI without sacrificing security.

Furthermore, digital credentials, considered "passports" for the digital economy, have a long validity period and rarely change. However, with regulators urging companies to adopt a mandatory 47-day credential renewal cycle, the large number of credentials requiring manual renewal in the past makes errors highly likely. Therefore, Palo Alto Networks released the Next Generation Trust Security Solution (NGTS), which transforms manual operations into a non-automated network management mechanism to avoid service interruptions and improve operational efficiency.