The Intellectual Property India on August 07, 2026, issued Guidelines for the Use of Artificial Intelligence in Patent Examination Procedures.
These Guidelines regulates the use of Artificial Intelligence in patent examination, aimed at supporting efficiency and quality while preserving confidentiality, accountability, consistency, and the Examiner/Controller's independent judgment. The Guidelines apply to AI use across tasks like screening, classification, search, translation support, drafting support, technical comparison, and knowledge retrieval, and they define key terms including AI, Generative AI, AI prompts, public vs. private AI tools, AI-assisted output, and prompt engineering. Public AI tools are described as widely accessible but trained on broad, generic datasets (limiting depth in specialized domains), while private AI tools operate in closed, proprietary environments with tighter access control, making them more suitable for sensitive data.
A substantial portion of the document catalogs limitations and risks of using AI in patent examination, stressing that these require thorough human oversight. Key risks include: profile sensitivity (AI outputs shaped by prior user interactions, affecting neutrality and predictability); the need for prompt discipline (well-structured prompts yield better results than vague ones); hallucination/fabrication (AI confidently producing non-existent prior art, fabricated judgments, or misrepresented citations, requiring independent verification against authoritative sources); omission of critical claim details; false pattern matching (surface-level keyword similarity mistaken for legal/technical relevance); loss of technical context during summarization or translation; classification and search drift; over- or under-inclusive interpretations; "black-box" opacity in reasoning; confidentiality risks from sharing unpublished material with external tools; subtle misguidance despite fluent-sounding output; incomplete or outdated training data (lacking real-time database access and recent prior art); bias in training data affecting jurisdictional/technological/linguistic representation; and difficulty handling complex patent language, drawings, and nuanced legal concepts like claim construction or inventive step.
Given these risks, the Guidelines emphasize that AI is meant to assist, not replace, the Examiner or Controller, who retains full responsibility for every official act involving AI use — this responsibility cannot be diluted or transferred. Any AI-assisted output influencing search or examination must undergo manual review and be adopted only after the officer is personally satisfied of its correctness, relevance, and appropriateness. The document also notes that the Patent Office's subscribed search databases already include built-in private AI tools optimized for patent search, offering better data security and contextual understanding than generic public tools. The final section (Section 4, referencing Annexures I and II) is set to elaborate on specific use cases, associated risks, and safeguards, along with a compliance checklist and declaration, though this content is not fully included in the excerpt.
