TITLE:
Authenticity Is Not Reliability: Algorithmic Facial Identification and Pakistan’s Law of Automated Evidence
AUTHORS:
Saad Khalid Khan, Farkhanda Khaleeq Khattak, Umarzish Noor ul Ain, Muhammad Afra Siab
KEYWORDS:
Algorithmic Facial Recognition, Article 46-A, Automated Information System, Silent Witness Doctrine, Test Identification Parade, Criminal Evidence, Pakistan
JOURNAL NAME:
Beijing Law Review,
Vol.17 No.4,
November
13,
2026
ABSTRACT: A court can be satisfied that a recording is genuine and unaltered without being satisfied that a computer has correctly identified the person in it. No reported Pakistani case has yet turned on an algorithmic identification, but Punjab’s deployed facial-recognition capability makes it a matter of time. The Qanun-e-Shahadat Order contains a complete chain for machine-generated evidence, inserted in 2002 and read in practice through Article 164: Article 46-A on relevance, Explanation 3 to Article 73 on class and the presumption of working order, Article 78-A on proof, and Article 59 on expert opinion about information systems, to which Zahir Zakir Jaffar adds a two-fold test for CCTV that it derives from Ahmed Omar Sheikh. Relevance, class, proof, and integrity are provided for; accuracy is not. The presumption of working order should be read as presuming that the machine was running, not that it was correct: on Millat Tractors, an Explanation is read by what it was intended to do and does, and this one classifies output and presumes that the system was operating; on Raja Khurram Ali Khan, the legal burden cannot be shifted to an accused unless the legislature by express terms commands otherwise. The missing requirement can be supplied from Article 59 with the gatekeeping principle, from the Silent Witness doctrine received in Zahir Zakir Jaffar and extended to a system that concludes rather than records, and from identification-parade doctrine. Drawing them together develops doctrine rather than restating a rule already in force. Because Pakistan tries criminal cases without juries, the argument concerns the foundation a court must be given, not a power to exclude. Three institutional routes in Punjab sharpen the problem, which is not wholly anticipatory: in review proceedings in Zahir Jaffar, the Supreme Court rejected an objection that forensic reports were inadmissible because their authors were not produced. Indian practice under Section 65B of the Evidence Act answered the same problem with a certificate, and so, in its own field, does Punjab’s Electronic Data Certificate; both secure integrity, neither accuracy. China yields lessons about classification and institutional separation, not a model of admissibility. We propose a four-stage framework and apply it to a worked example.