India proposes new security audit rules for AI models like OpenAI

India's nodal cybersecurity agency, CERT-In, has proposed a comprehensive security-audit framework for artificial intelligence systems that could subject major AI companies, including OpenAI and Anthropic, to much greater scrutiny. The draft guidelines focus on how models are developed, trained, and deployed across the country, particularly as many users search for details regarding India OpenAI operations and regulatory developments.

The Artificial Intelligence Security Auditing Guidelines arrive amid escalating concerns surrounding cybersecurity vulnerabilities linked to advanced artificial intelligence systems. Under this proposal, AI models and deployments would be classified into low, moderate, high, or critical risk categories, determining the level of oversight they receive throughout their operational lifecycle.

Understanding India OpenAI and New Audit Rules

The framework places specific emphasis on autonomous AI agents capable of performing independent tasks, accessing tools, and interacting with external systems. According to the proposed rules, such agents must operate under the principle of least privilege, meaning they should not automatically inherit permissions from users, tools, or other agents. Furthermore, high-risk agentic systems would be required to feature a dedicated kill switch mechanism allowing operators to halt operations when necessary.

The guidelines also mandate rigorous scrutiny of training data and development lifecycles. This includes verifying whether datasets are appropriately sourced, accounting for potential biases, and ensuring robust safeguards are maintained. Systems categorized as critical risk will face the most stringent requirements, such as continuous monitoring and independent security audits conducted by organizations empanelled specifically by CERT-In.

Audits will extend beyond conventional vulnerabilities to examine measures that mitigate AI hallucinations, as well as checking whether outputs contain harmful, biased, or copyright-protected materials. Additionally, the framework expects companies to utilize proper labelling or watermarking techniques to identify synthetic content depending on the context of its deployment.

AI-related security incidents must be reported to CERT-In within six hours under the proposed guidelines.

Why it matters

These proposed guidelines bring artificial intelligence systems directly under CERT-In's broader cybersecurity incident-response framework, creating strict accountability for developers and operators in India. By enforcing a six-hour reporting window for security incidents, continuous monitoring for critical systems, and strict operational controls like kill switches for high-risk agents, the framework aims to significantly mitigate AI-related risks and protect users from potential harms.