1. What are some of the biggest challenges organizations face today in establishing effective AI governance, particularly when AI systems are being deployed across critical infrastructure?

The biggest challenge is that AI governance is often treated as a technology or compliance issue, whereas it is actually a combination of business risk, cybersecurity, privacy, data governance and accountability. In critical infrastructure such as defence, BFSI, power, railways and government, organizations need visibility across the complete AI lifecycle from data and model development to deployment and decision-making. Based on my experience across defence, SOCs, State Data Centres and government infrastructure, I believe AI governance must integrate data lineage, model risk, cybersecurity, privacy, compliance and continuous monitoring rather than operate as a standalone function.

2. As organizations increasingly rely on AI-driven systems, how can they balance innovation with data privacy and security requirements?

Organizations should adopt Security-by-Design and Privacy-by-Design rather than trying to secure AI after deployment. AI should have access only to the data required for a specific purpose, based on identity, authorization and context. Techniques such as data minimization, encryption, masking, DLP, access controls and audit trails are essential. I believe the Zero Trust principle should extend to AI never assume the model, user, data or application is trustworthy; continuously verify before granting access or allowing actions.

3. What are the key data privacy risks associated with the use of AI in critical infrastructure, and what measures can organizations implement to mitigate them?

Major risks include data leakage through prompts, excessive data collection, unauthorized secondary use, inference of sensitive information, inappropriate retention and third-party AI exposure. Organizations need strong data classification, lineage, access control, encryption and retention policies. For critical infrastructure, every AI decision involving sensitive information should be traceable to its source data and purpose. India’s DPDP framework and sector-specific requirements should also be incorporated into the AI governance lifecycle.

4. From a cybersecurity perspective, what are the major vulnerabilities introduced by AI-enabled systems, and how should organizations strengthen their security frameworks to address these risks?

AI introduces new attack vectors such as prompt injection, data poisoning, model manipulation, adversarial attacks, sensitive-data extraction and AI supply-chain compromise. AI agents with excessive privileges are particularly concerning because one compromised agent could potentially access multiple enterprise systems. Organizations should therefore extend Zero Trust, least privilege, identity management, segmentation and behavioural monitoring to AI. Existing SIEM, SOAR, UEBA, NDR and threat-intelligence capabilities should also evolve to monitor AI-specific risks.

5. How can organizations improve their AI governance frameworks to ensure transparency, accountability, and responsible use of AI without slowing down business innovation?

AI governance should be risk-based rather than bureaucratic. A low-risk productivity assistant should not face the same approval process as an AI system supporting credit decisions, defence or critical infrastructure. Organizations should establish risk tiers with proportionate controls, clearly defined ownership and human accountability. This allows low-risk innovation to move quickly while high-risk AI receives deeper validation. Importantly, AI may support or automate decisions, but accountability must always remain with the organization and designated human stakeholders.

6. What role should continuous monitoring, risk assessment, and compliance play in protecting critical infrastructure from emerging AI and cyber threats?

Traditional compliance provides a point-in-time assessment, whereas AI and cyber threats evolve continuously. My experience with SOCs across government, defence and critical infrastructure demonstrates the importance of continuous visibility and assurance. Organizations should monitor AI models, agents, APIs, data access and abnormal behaviour alongside conventional network, endpoint and identity telemetry. SIEM, SOAR, UEBA and threat intelligence can provide this visibility. Simply put: risk assessment identifies what could go wrong; continuous monitoring identifies what is going wrong.

7. What practical solutions or best practices would you recommend to organizations looking to integrate AI while maintaining strong data protection and cyber resilience?

Organizations should begin with an AI inventory and risk classification, followed by strong data lineage and Zero Trust access controls. AI systems should undergo security testing and red teaming for prompt injection, data leakage, poisoning and privilege escalation. AI telemetry should be integrated into the SOC, while high-risk decisions should retain appropriate human oversight and fallback mechanisms. Organizations should also develop AI-specific incident-response and business-continuity playbooks. The objective should be to enable AI innovation while maintaining continuous security and privacy assurance.

8. Looking ahead, what improvements do you believe are most important for building a secure, privacy-conscious, and resilient AI ecosystem for critical infrastructure?

The future of AI security will require AI identity and access management, continuous model assurance, secure AI supply chains, privacy-enhancing technologies, AI red teaming and AI-native security operations. For critical infrastructure, collaboration between government, regulators, infrastructure operators, technology providers and cybersecurity organizations will also be essential. My view is that we need to move “From Zero Trust to Zero-Assumption AI” never assuming that the data, model, output or AI privileges are trustworthy. Critical AI decisions must be continuously validated, monitored and governed to create an ecosystem that is secure, privacy-conscious, resilient and accountable.