TRADE SECRETS v. ALGORITHMIC TRANSPARENCY IN AI: HOW INDIAN DATA PROTECTION AUTHORITIES CAN AUDIT PROPRIETARY LLMs WITHOUT VIOLATING IP

The rapid proliferation of Large Language Models in India's financial, healthcare, and public-service sectors has intensified a fundamental tension in information law: the right of AI developers to protect commercially sensitive model designs and training data as trade secrets, versus the public interest in auditing these systems for bias, privacy infringements, and accountability failures. The Digital Personal Data Protection Act, 2023 (hereinafter referred to as DPDP Act), together with the DPDP Rules, 2025, establishes an initial yet incomplete framework for algorithmic oversight. However, the absence of a dedicated trade secrets statute leaves intellectual property protection reliant on common law, contract, and equity. This article analyzes the doctrinal divisions between these competing imperatives, reviews international audit models, particularly the EU AI Act, 2024, and proposes a structured, IP-respecting audit architecture for Indian data protection authorities. The proposed framework incorporates black-box testing methodologies, tiered secrecy protocols, and regulatory sandboxes. It is argued that the perceived binary between disclosure and secrecy is misleading; robust audit frameworks can achieve regulatory accountability without necessitating disclosure of proprietary model weights, architectures, or training corpora. Keywords : Large Language Models, Algorithmic Transparency, Trade Secrets, DPDP Act 2023, EU AI Act.