Nat Natraj, Co- Founder & CEO, AccuKnox

Nat Natraj is a visionary leader in cybersecurity and AI-driven innovation, shaping cutting-edge solutions in threat detection, anomaly prevention, and runtime protection across cloud and enterprise environments. With a passion for advancing secure and ethical technology, he combines technical mastery with strategic insight to drive scalable, high-impact initiatives. His pioneering work safeguards sensitive data while empowering organizations, bridging innovation, responsibility, and transformative digital security.

In a recent interaction with M R Yuvatha, Senior Correspondent at siliconindia, Nat Natraj, shared his insights on ‘The Self-Evolving AI Aegis for Multi-Cloud AI & Hybrid Workloads’.

Intelligence is not fixed anymore, with the age of expanding clouds and mixed workloads. The Self-Evolving AI Aegis is a sentinel that thinks, adapts, and provides protection to multi-cloud environments. Beyond simple automation, it predicts, optimizes, and evolves, seamlessly integrating security, cost efficiency, and compliance into a single adaptive framework, ensuring intelligent, proactive, and resilient cloud operations.

AI Governance in Multi-Cloud Ecosystems

As organizations deploy AI across diverse cloud and hybrid environments, ensuring robust governance and security becomes a critical challenge. Addressing AI governance and security in hybrid and multi-cloud ecosystems is complex, requiring organizations to navigate people, process, technology, and strategy simultaneously. Responsible AI varies by context, in banking, for example, it demands fairness in algorithmic decisions to prevent biases across demographic segments. AI’s probabilistic nature adds further complexity, as outputs may vary for the same input, unlike deterministic software. Variations can arise from computational randomness, user biases, prompt injections, or malicious manipulations, which also pose security risks, including inadvertent exposure of sensitive or proprietary data.

Organizations must maintain strict data segregation across business units, implement multi-factor authentication, rotate access credentials, and enforce multi-level authorization workflows. Sensitivity differs by context an AutoCAD file may be critical in a chemical plant or architectural firm but trivial elsewhere necessitating governance structures that scale with AI’s far-reaching impact.

Given AI’s rapid growth, proactive and responsible measures are essential to manage its profound societal impact effectively.

To address these challenges, organizations must design ethical decision-making frameworks that balance AI-driven threat responses with human oversight, sandbox testing, and adherence to sensitive operations, cross-border data policies, and accountability principles. In India, IIT Madras has established a centre for Responsible AI to guide such initiatives.

Key approaches focus on four areas:

• People - right combination of technical and domain personnel who appreciate the business, technical and financial impacts of processes, workflows who can establish appropriate guardrails
• Process - process to address routine automations, escalations with manual checks to address false positives and false negatives, anomalies, configuration drifts
• Technology - Agentic workflows and relevant reporting dashboard, exception processing with pertinent access controls
• Governance/Oversight - Continuous Monitoring, Compliance, Escalation, Table Top Exercise

Also Read: The Invisible Risk of Untracked Data Movement in Enterprise Security

The Real Challenges of AI

Nowadays, integrating legacy systems into the cloud is not considered a major challenge, as migrating data from legacy ERP systems has already been widely addressed. What is unique to AI is the training process, as its effectiveness depends entirely on how it is trained. Models developed in one context, such as the US, may not perform accurately in India due to differences in training data.

For example, training a model for ICICI Bank requires analyzing decades of historical loan data including approved, denied, defaulted, and successfully repaid loans while accounting for anomalies such as the COVID-19 period. This training data forms the foundation for inferencing, where new loan applications, whether individual or corporate, are evaluated against the model to determine approval, denial, or specific loan covenants.

Accurate and context-specific training data is critical, as AI models are highly sensitive to input quality, following the principle of ‘garbage in, garbage out’ ensures reliable and actionable predictions.

Fostering a culture where every employee, not just IT, actively contributes to cybersecurity particularly in AI-driven environments is challenging, as people often resist change and cling to familiar routines. Adoption of new policies can be uncomfortable, requiring organizations to use a mix of incentives and penalties. A ‘carrot’ approach rewards employees who complete AI security competencies with recognition, bonuses, or awards, while a ‘stick’ approach imposes penalties for non-compliance, such as termination, as seen in Accenture’s mandatory AI certification policy. The most effective strategy balances both approaches, motivating self-driven employees while holding others accountable, aligned with the organization’s culture and personality.

The Human-Centric AI Imperative

In a democratic country like India, managing the impact of AI requires coordinated action at multiple levels central, state, local schools, communities, and across industries. For instance, banks can collaborate to share insights on effective training programs without exposing consumer data, recognizing that rural and metropolitan banks have different needs. While companies are making efforts, more proactive measures are necessary to ensure AI is embraced responsibly, balancing automation with fairness, ethics, and societal benefit.

Public awareness must highlight not just potential job loss but also opportunities for job creation. Like any tool, AI can be used positively or negatively, similar to precision knives in surgery or social media platforms. Its responsible use depends on individuals, organizations, societal frameworks, and government policies. Given AI’s rapid growth, responses cannot be casual, proactive and responsible measures are essential to manage its profound societal impact effectively.

Looking Ahead

Technological shifts are not just tools, they reshape lives. So, governments need to play a very important role in managing this tech transition.  Failing to act fairly, pro-active and insightful role can lead to mass-scale societal dislocations, social unrest (similar to what India experienced during the Bank Computer Automation in the late 1980s where there were massive union protests).

Some essential measures include:

• Training, education, certification by companies, college/university collaborations
• Re-skilling, job assistance, counseling, alternate min income for individuals whose entire industries have been eliminated (like voice transcription, entry level programmers)
• Legal, law enforcement oversight, timely hearing for abuse, danger to life - blackmail, impersonation, identity theft and sexual harassment.