Artificial intelligence (AI) is playing a bigger role in how cybersecurity is handled, especially as networks become more complex and data volumes continue to grow. Traditional approaches are finding it harder to keep up and AI is increasingly being used to process large amounts of information, spot patterns and support faster decision-making.
However, its role in security is not entirely straightforward. AI systems rely on constantly evolving data and operate across multiple layers of infrastructure, which makes their behaviour less predictable than conventional software. This changes how security itself needs to be approached.
The same capabilities that strengthen defence are also being used to make attacks more effective. As a result, the security landscape is becoming more difficult to read, with threats that are more targeted, more scalable and often harder to distinguish from normal activity. This shift is most visible in the way attacks themselves are evolving.
Evolving threat landscape
One of the clearest changes is in social engineering. Generative AI is enabling phishing messages, voice clones and deepfake content that are far more convincing than before. In India, this is already affecting sectors such as banking, telecom and digital payments, where trust and identity verification are central. Attackers are no longer relying on generic messaging; they are crafting communication that feels contextual and credible. In 2025, CERT-In advisories and bank alerts pointed to a rise in AI-driven phishing and voice scams targeting UPI users, often involving impersonation of officials or known contacts.
Voice cloning, in particular, is emerging as a serious concern. With minimal audio input, AI models can replicate voices with high accuracy. Reported cases across Indian enterprises have shown attackers using cloned voices of senior executives to push through urgent fund transfers or extract sensitive information, exposing weaknesses in voice-based authentication and approval systems.
Beyond social engineering, AI is also changing how technical attacks are executed. Vulnerability discovery is becoming faster, with AI models capable of scanning large codebases and identifying weaknesses more efficiently. This reduces the time between identifying a flaw and exploiting it. In distributed environments such as telecom networks, even small vulnerabilities can be replicated across multiple nodes, increasing the scale of impact.
Further, AI is influencing how malware behaves. Instead of relying on static code, malware can now adapt based on the environment it encounters, making it harder to detect using traditional methods. Polymorphic variants, which continuously modify their structure, are becoming more effective, reducing the reliability of signature-based detection.
Further, AI systems themselves are emerging as targets. Adversarial attacks, where inputs are subtly altered, can cause models to produce incorrect outputs. In telecom networks, this could lead to misclassification of traffic and allow malicious activity to pass through undetected. In fraud detection systems, it could result in suspicious transactions being missed. Instances of prompt manipulation in enterprise AI tools are also increasing, with attackers attempting to override safeguards or extract sensitive data.
Model poisoning adds another layer of risk. If compromised data enters training pipelines, it can gradually influence how models behave. This is particularly relevant in environments that rely on aggregated or externally sourced data sets, where maintaining data integrity is more difficult.
There is also growing exposure through the AI supply chain. Organisations increasingly depend on pre-trained models, open-source frameworks and third-party platforms. While this accelerates deployment, it introduces dependencies that are not always fully visible or controlled. A vulnerability in one component can become an entry point into the broader system.
What makes these threats difficult to manage is that they do not always present themselves as clear anomalies. These attacks are harder to isolate because they increasingly blend into normal system behaviour.
Securing AI systems
As AI becomes more embedded in core operations, securing it requires a shift in how cybersecurity is approached. It is no longer enough to focus on networks and endpoints alone. The attention needs to extend to models, data and environments in which they operate.
A key starting point is building security into AI systems from the outset. In many cases, safeguards are still being added after deployment, which leaves gaps. Organisations need to understand how models behave under unexpected conditions, define limits around outputs and account for failure scenarios early in the development process.
Data handling becomes equally important. Since AI systems depend on large volumes of data, the integrity of data pipelines directly affects how these systems perform. This means validating data sources, controlling access and continuously monitoring for inconsistencies or anomalies over time.
As AI systems are increasingly exposed through APIs and integrated across platforms, access control becomes more critical. Limiting who can interact with these systems, strengthening authentication and ensuring proper separation between workloads can reduce the risk of misuse.
Another challenge is visibility. Many organisations do not have a clear picture of where AI is being used, which models are deployed or what external components are involved. Without this visibility, it becomes difficult to assess risk or respond effectively when something goes wrong.
AI systems also require ongoing oversight because they evolve over time. Models are updated, retrained and fine-tuned, which introduces new points of risk. Securing these processes involves maintaining version control, validating updates and continuously monitoring for unexpected changes in behaviour.
The reliance on external components further complicates this landscape. Pre-trained models, open-source tools and third-party platforms are widely used, but they also introduce dependencies that need to be reviewed regularly. Security can no longer be limited to internal systems alone.
Regulatory considerations
As AI deployment scales, regulatory oversight is beginning to evolve in response. In India, the Digital Personal Data Protection (DPDP) framework marks a shift towards stricter accountability, with clearer expectations around consent, transparency and breach reporting. Organisations are now required to explicitly inform users about how their data is being used and to report breaches to both users and regulators. For sectors such as telecom, which handle large volumes of sensitive data, this raises the level of operational responsibility.
However, applying these principles to AI systems is not always straightforward. AI models rely on large and continuously evolving data sets, which do not always align with requirements such as purpose limitation and data minimisation. Data collected for one purpose is often reused for training or improving models, creating ambiguity around consent and usage boundaries.
This becomes more complex in telecom environments, where operators manage vast amounts of user data, while also using AI to enhance security. Bharti Airtel’s 2026 roll-out of an AI-based system to detect OTP-related fraud in real time and Vodafone Idea’s Vi Protect platform for spam and fraud detection illustrate how operators are using data not just for service delivery, but also for active risk mitigation.
Cross-border data flows add further complexity. AI development often depends on global infrastructure, while regulatory expectations around data localisation continue to evolve. Although the DPDP Rules provide more clarity, implementing these requirements remains challenging for organisations operating across multiple jurisdictions.
The framework also signals a stronger emphasis on enforcement. The establishment of the Data Protection Board of India, along with significant penalties for non-compliance, reinforces the need for organisations to treat data governance as a core operational priority, rather than just a procedural requirement.
Challenges
Despite the growing integration of AI across telecom networks and digital infrastructure, there remains a clear gap between deployment and preparedness. Organisations are moving quickly to scale AI, often prioritising efficiency and performance, while security considerations are addressed later.
There are structural challenges. AI security requires a different set of capabilities compared to traditional cybersecurity, yet many organisations are still relying on existing skill sets and frameworks. This creates gaps in areas such as model validation, monitoring and risk assessment.
Another constraint is the lack of clear and standardised governance practices. While organisations are beginning to adopt internal frameworks around responsible AI, these efforts are often fragmented and vary widely across sectors.
The way forward
Looking ahead, the integration of AI into business operations will continue to deepen. The key question is not whether AI adoption will expand, but how securely and responsibly it will be implemented.
One of the most important shifts will be towards security-by-design. This also includes testing models against manipulated inputs and edge cases before deployment, to ensure they do not fail in unpredictable ways. This means embedding safeguards across the entire AI life cycle, from data collection and model training to deployment and continuous updates. It also requires stronger mechanisms for validation, explainability and auditability, particularly in sectors handling critical infrastructure.
From a policy standpoint, India is still in a transitional phase. Frameworks provide a foundation, but AI-specific governance is still evolving. Going forward, greater clarity around accountability, risk classification and compliance requirements will be necessary.
For industry players, the focus will need to move towards integrated risk management. AI systems will have to be treated as core infrastructure, on par with networks and IT systems. This also calls for closer collaboration across telecom operators, technology providers and regulators, as risks increasingly span across interconnected ecosystems.
Talent and capability building will be another critical area. Managing AI-related risks requires specialised expertise that goes beyond conventional cybersecurity skills. Without the right capabilities, even well-designed systems may remain vulnerable.
Ultimately, organisations that will be best positioned are not those that adopt AI the fastest, but those that manage it most effectively. For India’s telecom ecosystem, this will mean aligning innovation with resilience, ensuring that the expansion of AI capabilities does not come at the cost of security, trust or long-term stability.
