Legal scholar Rohit Pradhan warns of 'prompt injection' attacks in e-filings that could manipulate AI tools used by Indian courts. Calls for judicial awareness, e-filing safeguards, and legal accountability frameworks to prevent covert influence.
Hidden AI Prompts Pose Judicial Risks
As Indian courts increasingly adopt AI for legal research, summarisation, and e-filing processing, a new threat—'prompt injection'—has emerged. These are concealed AI instructions embedded in pleadings that can manipulate AI outputs used by judges, potentially skewing case understanding or recommendations. Unlike hallucinations, these are deliberate exploitation attempts exploiting software vulnerabilities.
Rohit Pradhan, an IP litigator, highlights that e-filing formats—especially those incorporating metadata or hidden text—could be weaponised to alter AI-generated summaries or citations. Without sanitisation protocols, malicious actors could influence judicial cognition indirectly. While Indian jurisprudence has no precedent on this issue, procedural fairness under Article 14 demands proactive safeguards.
Pradhan urges courts to adopt content sanitisation standards, audit AI tools, and develop legal doctrines for accountability. As AI integration accelerates, integrity of judicial decision-making tools must be protected with technical and doctrinal solutions.
