TRACK-LLM is an AI Governance Testing Framework being developed by Amrita Vishwa Vidyapeetham in partnership with the Telecommunication Engineering Centre (TEC), Department of Telecommunications, Government of India, under the Safe & Trusted AI pillar of the IndiaAI Mission.
As Large Language Models (LLMs) become increasingly embedded in digital services, enterprise systems, education, healthcare, governance and everyday decision-making, the question is no longer only how capable an AI system is.
An equally important question is:
How do we know whether an AI system can be trusted in the context in which it is being used?
TRACK-LLM seeks to contribute towards answering this question by developing approaches for the governance testing of Large Language Models and their downstream applications.
Large Language Models are transforming the way people interact with technology.
LLMs now power conversational AI, intelligent assistants, enterprise applications, educational tools, knowledge systems, software-development platforms and increasingly complex AI agents.
Unlike conventional software, however, Large Language Models do not always produce identical outputs for identical types of tasks. Their responses can depend on the way a question is framed, the context supplied to the model, the information available to it and the environment in which it has been deployed.
They can generate highly fluent and convincing responses while still producing information that may be inaccurate, incomplete, inadequately supported or inappropriate for a particular context.
This creates a fundamental challenge for Responsible AI.
Traditional performance metrics alone may not sufficiently answer questions such as:
TRACK-LLM focuses on this emerging governance challenge.
TRACK-LLM stands for:
Transparency
Understanding and evaluating whether sufficient information is available about an AI system, its operation, capabilities and limitations.
Risk Assessment
Identifying and evaluating risks that may emerge when an LLM is integrated into a particular application, workflow or decision-making environment.
Context
Assessing whether the behaviour and outputs of the system appropriately account for the context in which the AI application is being used.
Knowledge
Examining the reliability and validation of knowledge communicated or generated through Large Language Model applications.
Together, these dimensions form the central idea behind TRACK-LLM:
evaluating an AI application not only for what a model can do, but also for how responsibly it can be deployed.
An LLM may perform extremely well on a standard benchmark and still present challenges when deployed in the real world.
A general-purpose language model, for example, could be integrated into:
The underlying model may be the same, but the risks associated with each application can be very different.
The consequences of an incorrect answer from a creative-writing assistant are fundamentally different from the consequences of incorrect information generated within a health, financial or public-service application.
This makes the downstream use case an important component of AI governance.
TRACK-LLM seeks to identify and address gaps in existing governance testing frameworks related specifically to LLM downstream use cases and deployment. The project was selected under the AI Governance Testing Framework theme of the Safe & Trusted AI pillar of the IndiaAI Mission.
Recent reporting based on information placed before Parliament describes the Amrita–TEC framework as evaluating Large Language Models across parameters that include:
Traceability
The ability to examine and track relevant information associated with the operation, inputs, outputs or decision pathways of an AI application.
Transparency
The degree to which important information about the AI system, its behaviour and its limitations can be understood by relevant stakeholders.
Risk Assessment
The systematic identification and assessment of potential risks arising from the deployment of a Large Language Model in a particular environment.
Contextual Understanding
Whether the system appropriately accounts for the context in which information is requested, interpreted or applied.
Knowledge Validation
The evaluation of the reliability and validity of knowledge communicated through an LLM-based system.
In August 2026, the AI Governance Testing Framework being developed by Amrita Vishwa Vidyapeetham and TEC was highlighted in reporting on information provided to Parliament on initiatives under the IndiaAI Mission.
The question surrounding Artificial Intelligence is rapidly evolving from:
“Can the model perform the task?”
to:
“Can the system be responsibly trusted to perform this task in this environment?”
This distinction becomes particularly important as Large Language Models move from experimental applications into operational systems.
Responsible deployment may require developers and organisations to consider:
Governance testing can help provide structured mechanisms for addressing such questions.
TRACK-LLM is one of the Responsible AI projects selected under the Safe & Trusted AI pillar of the IndiaAI Mission.
The Government of India approved the IndiaAI Mission in March 2024 with an outlay of ₹10,371.92 crore to build a comprehensive national AI ecosystem.
The Mission includes seven major pillars covering areas such as compute infrastructure, foundational models, datasets, AI applications, skills, startup financing and Safe & Trusted AI.
The Safe & Trusted AI pillar focuses specifically on developing indigenous mechanisms, tools, frameworks and guidelines that can support the responsible development and deployment of Artificial Intelligence.
In 2024, IndiaAI invited proposals for Responsible AI initiatives addressing a range of critical themes.
These included:
More than 2,000 proposals were received from academic institutions, startups, research organisations and civil-society organisations.
Following evaluation by a multi-stakeholder committee, eight Responsible AI projects were initially selected.
Under the theme AI Governance Testing Framework, the selected project was:
TRACK-LLM — Transparency, Risk Assessment, Context & Knowledge for Large Language Models
Selected Institutions:
Amrita Vishwa Vidyapeetham in partnership with the Telecommunication Engineering Centre
The selection was officially announced by the Ministry of Electronics and Information Technology in October 2024.
The Government continued to list TRACK-LLM among its Safe & Trusted AI initiatives in subsequent updates, including in 2026.
TRACK-LLM builds upon an earlier collaboration between Amrita Vishwa Vidyapeetham and the Telecommunication Engineering Centre (TEC).
TEC is the technical arm of the Department of Telecommunications, Ministry of Communications, Government of India, and is a recognised Standards Setting Organisation for telecommunications and related ICT systems in India.
On February 14, 2024, Amrita Vishwa Vidyapeetham and TEC signed a Memorandum of Understanding to collaborate in the area of Trustworthy and Responsible Artificial Intelligence Systems.
The collaboration was established to advance research around Responsible AI, with particular attention to bias, fairness and mechanisms for strengthening public trust in Artificial Intelligence technologies.
The Government of India stated that the collaboration aimed to develop tools to systematically assess bias risks and establish robust frameworks for evaluating the fairness and reliability of AI technologies. It also described the partnership as an effort to bridge academia and government and contribute to India’s leadership in AI.
The MoU was signed by Shri Avinash Agarwal, Deputy Director General, TEC, and Prof. Krishnashree Achuthan of Amrita Vishwa Vidyapeetham.
The development of TRACK-LLM represents an evolution of the Amrita–TEC collaboration as the AI landscape itself has rapidly changed.
When the MoU was signed in early 2024, Responsible AI research already included significant concerns around fairness, bias and reliability.
The rapid adoption of Generative AI and Large Language Models subsequently created an additional challenge.
LLMs are increasingly becoming components within larger applications rather than isolated models.
A contemporary AI application may combine:
The behaviour and risk profile of the resulting system may therefore depend on considerably more than the underlying foundation model.
TRACK-LLM addresses this emerging space of governance testing for downstream LLM applications.
Consider an organisation using the same Large Language Model in three different applications:
Application A: A writing assistant
Application B: An educational tutor
Application C: A health-information assistant
The technical foundation may be similar.
The level of risk is not.
Governance testing therefore needs to consider the application environment, intended users, information being processed and potential consequences of incorrect or inappropriate outputs.
This is one of the reasons governance frameworks for LLM applications are becoming increasingly important.
India represents one of the world’s largest and most diverse environments for the adoption of Artificial Intelligence.
The country combines:
Responsible AI systems deployed at this scale must operate across diverse contexts, populations and applications.
The Safe & Trusted AI initiative therefore places an emphasis on developing indigenous tools and governance mechanisms that reflect India’s requirements, datasets, challenges and opportunities.
TRACK-LLM contributes to this national effort through its specific focus on governance testing for Large Language Models.
As LLM adoption expands, governance testing may become relevant across a wide range of applications.
Education
AI tutors, learning assistants and content-generation tools need to provide information appropriate to the learner, educational context and subject domain.
Healthcare
AI-generated health information can carry significantly greater consequences if information is incorrect, incomplete or misunderstood.
Government and Public Services
Citizen-facing AI systems need appropriate mechanisms for accountability, reliability, transparency and risk management.
Financial Services
AI applications that influence financial decisions need strong safeguards around accuracy, appropriate communication and risk.
Enterprise AI
Organisations increasingly integrate LLMs with internal documents, databases and workflows, creating new questions around reliability, access, accountability and governance.
Software Development
Generative AI is increasingly involved in writing, reviewing and managing code, making the reliability and downstream implications of AI-generated outputs important.
These examples illustrate areas where AI governance is relevant. They should not be interpreted as claims that TRACK-LLM has already been formally validated or certified for deployment within each of these sectors.
Governance and innovation need not be opposing forces.
Effective AI governance can help organisations better understand:
Better mechanisms for evaluating AI systems can therefore support more informed and responsible innovation.
TRACK-LLM seeks to contribute to this emerging technological and governance infrastructure.
Amrita Vishwa Vidyapeetham and the Telecommunication Engineering Centre sign an MoU to collaborate in Trustworthy and Responsible Artificial Intelligence Systems.
The partnership focuses on responsible AI, bias assessment, fairness and the reliability of AI technologies.
TRACK-LLM is selected under the AI Governance Testing Framework theme of the Safe & Trusted AI pillar.
The project is one of eight Responsible AI projects selected following an Expression of Interest that received more than 2,000 proposals.
TRACK-LLM continues to be listed by the Government of India among projects being developed under the Safe & Trusted AI initiative.
The Government of India reports that the Safe & Trusted AI pillar has expanded to 13 Responsible AI projects as the IndiaAI Mission continues to strengthen India’s ecosystem for secure, fair and responsible AI.
Information provided in Parliament highlights the AI Governance Testing Framework under development by Amrita Vishwa Vidyapeetham and TEC.
The framework is described as evaluating Large Language Models across areas including traceability, transparency, risk assessment, contextual understanding and knowledge validation.
TRACK-LLM is currently under development.
Information on this page describes the publicly announced objectives and scope of the project.
TRACK-LLM should not presently be interpreted as providing regulatory certification or approval of any particular Large Language Model or AI application unless such a role is formally announced by the relevant Government agencies and project partners.
Research outputs, framework documentation, tools, publications and other resources will be added as they become publicly available.
The advancement of Artificial Intelligence is creating systems with extraordinary capabilities. But capability alone is not sufficient.
As AI becomes part of increasingly consequential applications, society also needs mechanisms to understand:
TRACK-LLM brings these questions into the process of evaluating Large Language Model applications.
Building the foundations for AI that is not only powerful, but responsible, transparent and trustworthy.
Press Information Bureau, Ministry of Communications — February 14, 2024
Telecommunication Engineering Centre and Amrita University sign MoU for collaboration in trustworthy and responsible Artificial Intelligence systems.
Amrita Vishwa Vidyapeetham — Original MoU Announcement
February 14, 2024
Amrita’s existing report on the collaboration between its Centre for Cybersecurity and the Telecommunication Engineering Centre.
ETGovernment — Media Coverage of Amrita–TEC Collaboration
February 15, 2024
Department of Telecom joins Amrita University for developing trustworthy, responsible AI.
Press Information Bureau, Ministry of Electronics & IT — October 16, 2024
IndiaAI announces the initial eight Responsible AI projects, including TRACK-LLM under the AI Governance Testing Framework theme.
Ministry of Electronics & Information Technology — Selected Responsible AI Projects
Official list identifying Amrita Vishwa Vidyapeetham and TEC for TRACK-LLM.
Press Information Bureau — July 27, 2026
Safe & Trusted AI pillar expands to 13 Responsible AI projects as part of the IndiaAI Mission.
Parliamentary Development — August 2026
Coverage of the Government’s update on the Amrita–TEC AI Governance Testing Framework and its evaluation parameters.
