Opinions expressed by Digital Journal contributors are their own.
“Autonomous AI systems introduce attack surfaces that traditional security controls never anticipated,” declares Prasidh Srikanth, Senior Director of Product Management at Palo Alto Networks. His assessment captures the fundamental challenge confronting global enterprises as they deploy artificial intelligence across production environments. The stakes escalate daily. Palo Alto Networks data reveals that generative AI traffic surged 890 percent throughout 2024, transforming experimental tools into mission-critical infrastructure practically overnight.
Prasidh spearheads technical innovation addressing this paradigm shift. His August 2024 launch of AI Access Security marked a pivotal moment for enterprise cybersecurity architecture. The platform tackles problems that emerged only recently, yet demand immediate solutions. Organizations adopting generative AI face unprecedented risks. Palo Alto Networks research found 99 percent of surveyed companies experienced at least one attack targeting their AI systems within the past year. Meanwhile, 75 percent now run AI in production environments, often without adequate governance frameworks.
Architecting runtime protection mechanisms
AI Access Security implements what Prasidh describes as “runtime enforcement for autonomous agents.” The approach differs fundamentally from conventional security tools that rely on signature matching or rule-based filtering. Autonomous AI systems make decisions dynamically, accessing data and executing actions without predetermined scripts. Protecting such systems requires security that operates at equivalent speeds and sophistication levels.
The platform provides visibility into more than 6000 generative AI applications proliferating across enterprise environments. Organizations often discover AI tools deployed without IT approval, creating what industry analysts term “shadow AI.” Gartner predicts 33 percent of enterprise applications will incorporate agentic AI by 2028, up from less than 1 percent in 2024. The acceleration leaves security teams scrambling to establish oversight before vulnerabilities get exploited.
Prasidh’s technical framework combines several innovations. Real-time telemetry captures granular details about AI system behavior, including which applications users access, what data flows through those applications and how AI agents interact with enterprise resources. The visibility enables security teams to establish baselines for normal behavior and detect anomalous patterns indicating potential threats.
Policy engines translate organizational security requirements into enforceable controls. Administrators define rules specifying which users can access particular AI applications, what types of data those applications may process and under what circumstances automated blocks should trigger. The system enforces these policies continuously, intervening before sensitive information reaches unauthorized destinations.
Solving the sensitive data challenge
Data loss prevention for AI systems presents unique technical challenges. Traditional DLP solutions monitor structured data flows between known endpoints. AI applications introduce unpredictability. Users might paste sensitive content into chatbot interfaces without realizing they’ve transmitted confidential information to external language models. Those models may retain data for training purposes, creating persistent exposure risks.
Prasidh’s approach employs machine learning-powered classification that identifies sensitive information across 1000’s of predefined categories. The system recognizes patterns indicating financial data, personal identifiable information, intellectual property and regulated content. Natural language processing algorithms understand contextual nuances that rule-based systems miss. They detect when seemingly innocuous text actually contains sensitive details requiring protection.
Accenture research indicates 57 percent of organizations express concern about data poisoning in generative AI deployments. The 2025 EchoLeak exploit against Microsoft Copilot demonstrated how engineered prompts could trigger automatic data exfiltration without user interaction. The AI Access Security platform implements defenses against such attack vectors through behavioral analysis that identifies suspicious prompt patterns before they execute.
The technical architecture integrates with cloud access security broker capabilities, creating unified enforcement across SaaS applications, browser extensions and endpoint devices. McKinsey projects agentic AI could unlock between $2.6 trillion and $4.4 trillion in annual value across more than 60 enterprise use cases by 2028. Realizing that value requires security architectures that enable adoption rather than block it entirely.
Patent-backed innovation and research contributions
Prasidh draws on years of research in distributed systems and security. He holds a U.S. patent US10803257B2 covering secure communication and access control, tackling the challenge of verifying entities across complex networks without slowing performance. His peer-reviewed work includes studies on cryptographic protocols for resource-constrained devices—a line of research that now informs how he helps develop products that protect autonomous AI agents. By connecting theoretical research to practical implementation, Prasidh has built a framework that balances security, usability, and adaptability in enterprise AI systems.
Industry validation and market recognition
AI Access Security received the 2024 Tech Ascension Award for Best AI/ML Powered Solution. The recognition validates the platform’s technical approach and market relevance. Industry analysts, including Forrester and Gartner, increasingly emphasize the need for specialized AI governance tools. Traditional security categories like firewalls, endpoint protection and identity management address different threat models. AI systems demand purpose-built controls acknowledging their unique characteristics.
Prasidh has presented this work at conferences organized by SANS, the leading cybersecurity training organization and at UC Berkeley. His talks explore emerging attack vectors and defensive architectures. He appeared on the Heavy Networking podcast discussing modern Data Security implementations, providing technical depth appreciated by practitioner audiences.
Strategic partnerships extend the platform’s reach further. Palo Alto Networks maintains deep integrations with OpenAI, Glean, Salesforce, ServiceNow, Google Workspace, Box and Atlassian, enabling AI Access Security to protect data across diverse enterprise environments.
Addressing agentic AI threat landscape
Autonomous AI agents present distinctive security challenges that Prasidh’s work directly tackles. Unlike traditional applications that follow predefined logic paths, agentic systems plan actions, use tools, maintain memory and adapt behavior based on observations. The autonomy enables powerful capabilities but introduces vulnerabilities that attackers can exploit.
Memory poisoning represents one threat category where malicious actors corrupt an agent’s knowledge base with misleading information. Subsequent decisions based on poisoned memory produce incorrect outcomes favoring the attacker’s objectives. Prasidh’s platform implements integrity checking for agent memory stores, validating information against trusted sources before allowing critical decisions.
Prompt injection attacks attempt to manipulate AI behavior through carefully crafted inputs that override intended instructions. The 2025 EchoLeak vulnerability demonstrated the severity of this threat class. Prasidh’s defenses employ multi-layered filtering that examines prompts for suspicious patterns while maintaining usability for legitimate queries. The system learns from attempted exploits, continuously updating detection algorithms.
Tool misuse occurs when attackers manipulate agents into using their capabilities maliciously. An AI agent with database access might be tricked into extracting sensitive records. One with code execution privileges could be coerced into launching unauthorized processes. Prasidh’s framework enforces least-privilege principles, granting agents only the minimum necessary permissions while logging all actions for audit purposes.
Technical methodology and implementation
Prasidh developed what he terms “persona-driven Shadow IT visibility” as part of the technical solution. The methodology classifies applications based on user roles, access patterns and organizational risk profiles. Traditional approaches monitor devices or network traffic without understanding the business context. Persona-based classification incorporates organizational structure, enabling more nuanced policy enforcement.
A financial analyst accessing AI tools for market research represents a different risk than a product manager querying customer data. The system recognizes these contextual differences and applies appropriate controls. Role-based access control models extend into AI governance, something earlier security architectures hadn’t contemplated.
Integration represents another technical achievement. Prasidh’s team unified Data Loss Prevention, Cloud Access Security Broker and cloud-native security telemetry into cohesive workflows. Traditional implementations treat these capabilities as separate point solutions requiring manual coordination. The convergence model automates detection, prevention and remediation across platforms. Threats identified in one component trigger coordinated responses throughout the architecture.
Performance optimization ensures security controls don’t degrade user experience. AI applications demand low latency to maintain conversational flows. Prasidh’s architecture implements inline inspection without introducing noticeable delays. The technical approach involves distributed processing where policy evaluation occurs at network edges rather than centralized chokepoints.
Forward trajectory and evolving threats
The AI security market reached $30.92 billion in 2025, and analysts forecast growth to $86.34 billion by 2030, reflecting a compound annual growth rate of 22.8 percent. The expansion stems from recognition that AI systems require specialized protection. Asia-Pacific markets project a 24.1 percent compound annual growth as digital transformation initiatives accelerate across emerging economies.
Banking and financial services sectors held 28.4 percent of the AI cybersecurity market in 2024, driven by strict compliance mandates and valuable data protection requirements. Prasidh’s solutions serve heavily regulated industries where security failures carry catastrophic consequences. The platforms must satisfy auditors and regulators while enabling business operations.
Prasidh anticipates continued evolution in the threat landscape and defensive technologies. “Adversaries adopt AI faster than defenders in many cases,” he observes. “We’re building systems that detect and respond to machine-speed attacks without human intervention for routine threats while escalating novel situations requiring expert judgment.”
His vision emphasizes autonomous security operations where AI-powered defenses protect AI-powered applications. The recursive nature reflects broader trends toward intelligent systems managing other intelligent systems. Organizations adopting this model can scale security operations without proportionally expanding staff. The approach addresses chronic talent shortages plaguing cybersecurity. Platforms incorporating automated triage, no-code automation and managed service bundles allow organizations to operate with leaner teams.
Prasidh’s technical innovations position enterprises to harness transformative AI capabilities while managing inherent risks. Whether through runtime enforcement, sensitive data protection, or autonomous threat response, his work addresses challenges that barely existed five years ago yet now dominate security priorities. The pace of change shows no signs of abating. Organizations implementing robust AI governance frameworks today position themselves for competitive advantage in an increasingly autonomous technological landscape.
