Digital evidence and presumption of criminality: ai-generated evidence under the u.p. goondas act, 1970

This article examines how AI-generated digital evidence such as deepfake videos, synthetic audio, facial-recognition outputs and metadata could influence preventive criminal proceedings under the Uttar Pradesh Control of Goondas Act, 1970 (U.P. Goondas Act). We first outline the Act’s framework: notice, hearing, externment/detention orders, appeal, etc. and the definition of “Goonda”. We then summarise relevant provisions of the new BharatiyaSakshyaAdhiniyam, 2023, especially Sections 62–63 on electronic records, which require strict certification (including hash values) for AI-generated media. Notably, BSA’s rules apply only in judicial proceedings, whereas Goonda Act proceedings are executive/quasi-judicial; Section 8 of the Goonda Act explicitly allows any probative evidence, setting aside ordinary Evidence Act rules. We then describe types of AI evidence like deepfake images/videos, voice clones, algorithmic ID, metadata logs, etc. and their inherent challenges: authenticity, chain-of-custody, bias, explainability. We analyse how reliance on such evidence might lead authorities to deem someone a “habitual offender” or threat under the Goonda Act, potentially creating a de facto presumption of criminality unless safeguards are imposed. We further review recent case-law (2023–2026) on digital evidence and preventive orders: for example, Ramji Pandey v. State of U.P. (1981), an Allahabad HC Full Bench insisted that a Goonda Act notice must state the general nature of allegations. More recently, the Supreme Court in State of Telangana v. Roshini Devi (2026) struck down a detention order under Telangana’s “Goonda Act” because it merely recited offences without showing a specific public order threat. Courts have emphasised that preventive detention cannot override bail rights without concrete evidence of danger, distinguishing “public order” from mere “law and order” situations. No Indian court has directly ruled on deepfake evidence in a Goonda context. Still, recent Bombay and Delhi High Court injunctions, e.g. Arijit Singh v. Codible Bombay HC 2024, Jackie Shroff v. Peppy Store (Delhi HC 2024), Preity Zinta deepfake case (Bombay HC 2026), show courts are aware of AI misuse, albeit in personality rights cases. We draw on comparative sources (U.S. Federal Rules, European AI proposals) to suggest safeguards: for instance, judicially vetting AI evidence, requiring disclosure of AI methods, maintaining a rigorous chain of custody (hashes, logs), and ensuring expert validation and cross-examination rights. The conclusion recommends policy steps legislative guidance or rules on AI evidence, mandatory audit trails, and caution in Goonda proceedings to protect liberty and due process.