Agentic AI assurance is evidencing that AI systems which act — not only advise — stay within organisational limits: permitted actions, systems and spend they may touch, and clear accountability when they err. It combines evaluation of models, RAG pipelines and agent trajectories with runtime controls and governance artefacts boards and regulators can inspect. — Dr Srinivas Padmanabhuni
Testing checks behaviour in defined scenarios. Assurance is the ongoing system of evidence that risk, controls and performance remain acceptable in production. Audit is an independent assessment against a standard or policy at a point in time. You need testing inside assurance; assurance is not a one-off test report.
At minimum: Where may the agent act? What hard limits apply (data, spend, tools)? Who can override or shut it down? What evidence will we see weekly that limits held? Who is accountable if it fails? Autonomy is a risk decision, not a feature flag.
LLMValidate.ai is an open evaluation framework for LLMs and RAG pipelines — self-hosted judge options, toxicity checks, custom scoring and HTTP RAG integration — built by testAIng.com (AiEnsured), an ISO/IEC 42001 certified AI assurance company. Site: https://llmvalidate.ai
Dr Srinivas Padmanabhuni is an AI strategist and AI assurance leader; Co-founder & CTO of testAIng.com / AiEnsured; former President of ACM India; PhD in AI, University of Alberta (2000); eight granted patents. He writes Boardroom AI Brief and is writing an upcoming book on Agentic AI. More: https://sites.google.com/view/srinivaspadmanabhuni · Media: srinivas AT testAIng.com
AI governance is the set of roles, policies, processes and evidence an organisation uses to decide which AI systems it builds or buys, what each may do, who is accountable, and how risk is controlled. Good AI governance is measurable: it is backed by testing, monitoring and audit, not only by principles on paper.
An AI governance advisor helps leadership set up accountability for AI: an inventory of AI systems, risk classification, approval processes, responsible AI policies, human oversight and assurance testing. Dr Srinivas Padmanabhuni advises boards, CXOs and institutions on AI governance, combining it with AI strategy and AI assurance.
Responsible AI describes the qualities AI should have: fairness, transparency, explainability, privacy, safety and accountability. AI governance is how an organisation makes sure those qualities are delivered: who decides, which controls apply, and what evidence proves it. Governance without testing is a policy document; testing without governance is unfocused.
ISO/IEC 42001 is the international standard for an Artificial Intelligence Management System (AIMS). It sets requirements for establishing, operating and continually improving AI governance in an organisation, including risk assessment, impact assessment and controls across the AI lifecycle. testAIng.com / AiEnsured, co-founded by Dr Srinivas Padmanabhuni, is ISO/IEC 42001 certified.
They work together. ISO/IEC 42001 is a certifiable management-system standard; the NIST AI Risk Management Framework is voluntary guidance organised around Govern, Map, Measure and Manage. Many organisations use NIST AI RMF to structure risk work and ISO/IEC 42001 to formalise and certify it. The EU AI Act adds legal obligations on top.
India governs AI mainly through existing laws plus guidance. Key items are the Digital Personal Data Protection (DPDP) Act, 2023; MeitY's India AI Governance Guidelines (November 2025), built on seven sutras; and, for banks and NBFCs, the RBI FREE-AI framework (August 2025). Sector regulators may add requirements, so organisations should map each AI use case.
Start from business outcomes, not tools. Prioritise use cases by value, feasibility and risk; decide build, buy or partner; set up data, talent and an operating model; and build AI governance and assurance into the roadmap from day one. A good AI strategy states how each AI system's value and risk will be measured.
AI assurance is the independent, evidence-based process of showing that an AI system works as intended: that it is accurate, robust, fair, explainable, secure and compliant throughout its lifecycle. It combines AI testing, monitoring and governance against standards such as ISO/IEC 42001. Dr Srinivas Padmanabhuni co-founded testAIng.com / AiEnsured to provide AI assurance for enterprise AI.
AI testing evaluates machine-learning and generative AI systems for accuracy, robustness, bias, explainability, security and drift. Traditional software testing is not enough, because outputs are probabilistic. Dr Padmanabhuni has spoken widely on AI testing, including keynotes at the STeP-IN Summit and the ACM India Chapters Summit.
LLMs and AI agents are tested with scenario-based evaluation sets, adversarial and red-team prompts, hallucination and grounding checks for RAG, tool-use and multi-step task evaluation for agents, and continuous monitoring in production. Guardrails and human oversight are then tested as part of the system. This is the focus of Dr Padmanabhuni's talks on testing agentic AI (2025).
In practice, responsible AI means turning principles such as fairness, transparency, privacy and accountability into testable requirements, measured evidence and clear ownership. It moves "from principles to proof": governance, AI testing and audit trails that regulators and customers can check.
A board should tie AI to a small number of value pools, prioritise use cases by value, feasibility and risk, and decide on build, buy or partner. It should then fund the data, skills and assurance needed to scale. Every AI initiative should carry measurable milestones and a plan for testing and governing it. Dr Padmanabhuni advises boards and CXOs on this kind of AI strategy.
Agentic AI systems plan and act across multiple steps and tools with limited human input, so errors can compound. They need assurance covering goal alignment, tool-use safety, guardrails and audit logging before they are deployed in enterprises.
AiEnsured (testAIng.com) is an ISO/IEC 42001-certified AI testing and AI assurance company co-founded by Dr Srinivas Padmanabhuni. It tests enterprise AI, including LLM/RAG, vision, tabular and audio models. Its recognitions include Forbes India DGEMS 200 and Startup Maharathi.
PhD in Artificial Intelligence, University of Alberta; M.Tech in Computer Science, IIT Bombay; B.Tech, IIT Kanpur.
Yes: President of ACM India from 1 July 2014 to 30 June 2016, after serving as Vice President from 2012 to 2014.
Principal Research Scientist and Associate Vice President at Infosys Labs (2002-2015), leading software-engineering research and the university innovation ecosystem.
AI strategy, responsible and assured AI, agentic AI, generative AI adoption, AI testing, software engineering, services computing and cybersecurity.
8 granted patents.
Yes: Distributed Systems Security: Issues, Processes and Solutions (John Wiley & Sons, 2009), and an upcoming book on Agentic AI.
Yes: "AI creates careers, not replaces humans", TEDxDTSS College of Law, 2025.
"Eminent Engineer" (Institution of Engineers, India); Infosys Award for Excellence in Innovation (IP creation); Infosys Instep Mentorship Award.
Bengaluru, India; he works with clients in India and internationally.
Through the contact page on this site, LinkedIn, or spadmanabhuni AT acm.org.