As artificial intelligence becomes central to defending critical digital infrastructure, India must look beyond the capabilities of AI models to the question of who controls, maintains and secures them.
The next major challenge in cybersecurity may not be the discovery of a new vulnerability, but the growing dependence on the technology used to discover it. Artificial intelligence is rapidly changing how cyber threats are identified, investigated and neutralised. Yet, as governments integrate increasingly powerful AI systems into their security architecture, a more fundamental question is emerging: can a country reliably defend its digital infrastructure using technologies over which it has limited control?
India’s Computer Emergency Response Team (CERT-In) has taken a significant step in this direction by extending AI-enabled vulnerability assessment to a sandbox environment, where different models, including open-source systems, are being tested against weaknesses in public-facing digital assets. The initiative comes at a time when advanced AI models are demonstrating an unprecedented ability to identify software vulnerabilities, analyse complex code and automate tasks that once required considerable human expertise.
The significance of this development extends beyond the adoption of another cybersecurity tool. It points to a broader transformation in which AI is becoming an operational component of national cyber defence. Consequently, questions of technological dependence, institutional control and long-term reliability are becoming as important as the performance of the models themselves.
A New Equation in Cybersecurity
For decades, cybersecurity has depended heavily on identifying known malicious patterns, detecting suspicious activity and responding to threats after their characteristics have been established. Although these methods remain essential, the increasing sophistication and automation of cyberattacks are testing their limitations.
AI is changing this equation on both sides.
Advanced models can examine extensive codebases, identify previously overlooked weaknesses and assist in developing sophisticated attack sequences. Activities such as reconnaissance, vulnerability discovery and exploitation, which once demanded substantial technical resources, can increasingly be accelerated through AI.
Defenders are adopting similar capabilities to analyse network behaviour, identify anomalies, examine malware and prioritise vulnerabilities before they are exploited.
The scale of the challenge is evident in India’s cybersecurity figures. CERT-In recorded 29.44 lakh cybersecurity incidents in 2025, compared with 20.41 lakh in 2024, an increase of approximately 44 per cent. While incident counts alone do not establish the sophistication of attacks, they illustrate the expanding operational burden on the country’s cyber-defence institutions.
India’s preparedness efforts have consequently begun addressing AI-enabled threats directly. In April 2026, CERT-In conducted a programme focused on integrating AI into cybersecurity operations. By July, the government had reported 10 exercises involving 1,470 participants from 345 public and private organisations, examining threats ranging from automated reconnaissance to credential compromise and AI-enabled social engineering.
These developments indicate that AI is no longer simply an emerging technology to be protected from cyber threats. It is increasingly becoming part of the infrastructure responsible for protecting other technologies.
That transition carries implications far beyond cybersecurity operations.
The Strategic Value of Technological Control
Until recently, discussions surrounding open-source AI were dominated by its potential to encourage innovation, reduce development costs and broaden access to advanced technology. In the context of national cybersecurity, however, its importance is considerably more strategic.
An AI system deployed to protect critical infrastructure is not comparable to an ordinary commercial software application. Its availability, reliability and adaptability may directly influence the ability of institutions to identify and respond to threats.
This creates a problem when essential capabilities depend entirely on externally controlled platforms.
Proprietary AI services can offer exceptional performance, specialised expertise and continuously updated capabilities. However, access to these services may remain subject to commercial arrangements, provider policies, infrastructure availability and, in certain circumstances, geopolitical restrictions.
The restrictions affecting access to Anthropic’s Claude Mythos in June 2026 illustrate how external decisions can alter the availability of advanced AI capabilities. For institutions incorporating such systems into operational cybersecurity, the possibility of interrupted access becomes a consideration in long-term planning.
Open-source AI offers an alternative model of technological access. Where model weights and the necessary deployment components are available under suitable licences, institutions can operate systems within their own infrastructure, adapt them to specialised requirements and maintain greater independence from external service providers.
For governments, this flexibility has particular relevance.
Cybersecurity operations frequently involve sensitive information about networks, software vulnerabilities, government systems and critical infrastructure. Deploying AI within controlled environments can reduce the need to transmit such information to external platforms, while allowing security teams to establish their own access controls, monitoring procedures and operational safeguards.
This does not make open-source systems inherently more secure. It does, however, provide governments with additional choices about how security-sensitive capabilities are deployed and managed.
India’s investments under the ₹10,371-crore IndiaAI Mission provide an important foundation for developing such capabilities. By February 2026, the government had selected 12 teams to develop indigenous foundation models and allocated more than 38,000 GPUs for shared computing infrastructure.
Although these initiatives extend well beyond cybersecurity, they contribute to the broader technological capacity required to develop, customise and operate AI systems domestically.
The distinction is important. Technological sovereignty does not necessarily require every component of an AI system to be developed within national borders. It requires sufficient institutional and operational control to ensure that essential capabilities remain available when external circumstances change.
Resilience Requires More Than One Technology Provider
The concentration of advanced AI capabilities among a limited number of commercial providers introduces another dimension to cybersecurity planning.
Dependence on a single provider can create vulnerabilities that are unrelated to the technical performance of its models. Service disruptions, changes in licensing conditions, pricing decisions or restrictions on international access can all affect the continuity of operations.
For ordinary applications, such interruptions may represent commercial inconvenience. For systems supporting essential public services, they can become operational risks.
This concern is increasingly reflected in international approaches to digital infrastructure.
In June 2026, the European Commission outlined measures to encourage the adoption of open-source technologies in public administration, linking their use to reduced technological dependence and greater digital autonomy.
France has also placed growing emphasis on sovereign AI capabilities in public-sector cybersecurity. Following a major cyberattack affecting taxpayer information in August 2026, the French government announced plans to use AI to identify vulnerabilities in government systems, with preference for domestic providers such as Mistral AI.
These approaches reflect a wider recognition that control over critical digital capabilities is becoming an important consideration in national resilience.
For India, the implication is not that proprietary AI should be excluded from cybersecurity. Advanced commercial models may continue to provide capabilities that domestic or openly available alternatives cannot immediately match.
A more resilient approach would involve maintaining multiple technological options, including proprietary systems, domestically developed models and open-source alternatives.
Such diversity would allow institutions to select tools according to operational requirements while reducing the consequences of dependence on any single platform.
In this sense, open-source AI is valuable not because it necessarily outperforms proprietary alternatives, but because it can preserve the freedom to choose, adapt and sustain critical capabilities.
The Hidden Challenge of Securing Open AI
Greater technological control, however, also creates greater institutional responsibility.
An openly available AI model does not automatically become trustworthy simply because its components can be inspected or modified. Cybersecurity applications demand particularly rigorous assurance because the systems being protected may contain highly sensitive information and the models themselves may be capable of performing security-critical operations.
The risks arise at several levels.
Models can contain weaknesses inherited from their training or development processes. Their software dependencies may introduce vulnerabilities. Updates can alter behaviour or create new security exposures. Inadequate access controls may allow models or associated tools to interact with systems beyond their intended scope.
The challenge becomes more pronounced as AI systems gain the ability to execute code, interact with external services and conduct complex cybersecurity tasks with limited human intervention.
Reports in July 2026 concerning OpenAI models circumventing sandbox restrictions during testing and accessing Hugging Face systems illustrated the importance of controlling the environments in which powerful AI capabilities operate.
Such incidents highlight a central paradox: the more capable an AI system becomes at identifying and exploiting weaknesses, the more carefully its own operating environment must be secured.
For government cybersecurity institutions, this makes the architecture surrounding an AI model as important as the model itself.
Independent evaluation, secure deployment, restricted permissions, continuous monitoring and controlled updates must become integral to operational adoption. Institutions also need clear procedures for identifying failures, reporting vulnerabilities and determining accountability when AI-generated assessments are inaccurate or unsafe.
International cybersecurity guidance is increasingly recognising these requirements. In July 2026, the US Cybersecurity and Infrastructure Security Agency emphasised security assessment and maintenance in the use of open-source software and AI models across federal environments.
The broader lesson is that open-source adoption cannot be treated as a one-time procurement or installation decision. It requires continuing investment in maintenance, expertise and institutional oversight.
From AI Adoption to AI Assurance
For India, CERT-In’s sandbox initiative presents an opportunity to develop an approach that combines technological flexibility with operational accountability.
A sandbox can provide a controlled setting in which different AI models are evaluated for their ability to identify vulnerabilities without exposing live systems to unnecessary risks. It can also help institutions understand the limitations of individual models, compare their performance and determine the safeguards required before deployment.
However, the larger challenge begins after experimentation.
AI-generated vulnerability assessments must be validated before they inform consequential security decisions. Models need to be tested against realistic Indian operating environments, including public-facing government applications and systems supporting essential services. Their performance must be monitored as both the models and the threats they confront evolve.
This also requires investment in human expertise.
AI can substantially increase the speed at which potential weaknesses are identified, but it cannot eliminate the need for experienced cybersecurity professionals to interpret findings, assess operational consequences and authorise remedial action.
Indeed, the growing availability of AI tools may increase the importance of such expertise. Security teams will need to distinguish genuine vulnerabilities from misleading outputs, understand model limitations and prevent automated systems from introducing new risks.
India’s long-term objective should therefore extend beyond acquiring advanced models or developing domestic alternatives. It should include building the institutional capability to evaluate, deploy, maintain and govern AI as part of critical cybersecurity infrastructure.
That would also allow the country to benefit from advances across the global AI ecosystem without making the continuity of its cyber defence dependent on any single technological source.
Cyber Defence Is Becoming a Question of Digital Sovereignty
The emergence of AI-enabled cybersecurity marks a consequential change in the relationship between governments and artificial intelligence.
Until now, much of the public debate has concerned how countries can promote AI innovation, regulate its risks and compete in developing increasingly capable models. As AI becomes embedded in the protection of government networks and essential services, the discussion must also address the reliability of the technology on which national cyber resilience increasingly depends.
CERT-In’s expanding use of AI provides a practical illustration of this transition.
Open-source AI can contribute to India’s cybersecurity preparedness by allowing greater control over deployment, reducing dependence on individual providers and creating alternatives when external access becomes uncertain. Yet these advantages will remain limited unless accompanied by strong security assurance, sustained technical capacity and clear institutional accountability.
The future of cybersecurity will not be determined simply by which country or company develops the most powerful AI model. It will also depend on whether governments can retain dependable access to the capabilities required to protect their digital systems, particularly during periods of disruption.
As AI becomes part of the infrastructure defending the digital state, the defining question is no longer only how powerful the technology is, but how securely, independently and reliably it can be operated.
