Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Hur potentiellt AI-genererad bevisning påverkar beviskravet – En studie om robusthet och sannolikhet

Wallanger, Jakob LU (2026) JURM02 20261
Department of Law
Faculty of Law
Abstract
This thesis examines how the rapid advancement of artificial intelligence, with a specific focus on generated imagery, impacts evidence and the duty to inve-stigate within Swedish criminal procedure. Grounded in the principles of free admissibility of evidence and the free evaluation of evidence under Chapter 35, Section 1 of the Swedish Code of Judicial Procedure, the study analyzes how courts evaluate, or ought to evaluate, defense claims that digital evidence has been manipulated.
The study combines a legal dogmatic method with two interdisciplinary frameworks: Probabilistic evidence theory (the Bayesian method) and law and economics (cost-benefit analysis) Through the concept of robustness, the the-sis investigates whether an... (More)
This thesis examines how the rapid advancement of artificial intelligence, with a specific focus on generated imagery, impacts evidence and the duty to inve-stigate within Swedish criminal procedure. Grounded in the principles of free admissibility of evidence and the free evaluation of evidence under Chapter 35, Section 1 of the Swedish Code of Judicial Procedure, the study analyzes how courts evaluate, or ought to evaluate, defense claims that digital evidence has been manipulated.
The study combines a legal dogmatic method with two interdisciplinary frameworks: Probabilistic evidence theory (the Bayesian method) and law and economics (cost-benefit analysis) Through the concept of robustness, the the-sis investigates whether an investigation can be deemed sufficient to support a conviction when facing a general or specific risk of digital manipulation. An investigation lacks robustness if additional, hypothetical investigation measu-res could reasonably alter the strength of the existing evidence in favor of the defendant.
The thesis applies these theoretical models to contemporary Swedish lower court case law. The analyzis demonstrates that an objection regarding AI ma-nipulation does not diminish the aggregate probative value if the remaining investigation is sufficiently robust. In such scenarios, the objection can be dismissed as an ex post fabrication. Conversely, if the digital evidence i central to the prosecution’s case and the remaining investigation lacks independent corroboration, forensic authentication becomes essential to satisfy the standard of proof.
Lastly, the thesis establishes an economic benchmark for the prosecutor’s duty to investigate during the preliminary inquiry. Given that state resources are finite and a routine authentication of all digital evidence would overwhelm the National Forensic Centre, forensic verifications must be constrained by cost-benefit parameters. An investigation is justified only when the potential of the verification to alter the outcome of the case (switch-ability) is real, which is determined by balancing the centrality of the evidence against how its digital and physical handling has been documented. (Less)
Abstract (Swedish)
Detta arbete undersöker hur den snabba utvecklingen av artificiell intelligens, med särskilt fokus på genererade bilder, påverkar bevisvärderingen och utred-ningsansvaret i svenska brottmål. Med utgångspunkt i principerna om fri be-visföring och fri bevisvärdering i 35 kap. 1 § rättegångsbalken analyseras hur domstolarna hanterar, eller bör hantera, invändningar från den tilltalade om att digital bevisning har manipulerats.
Arbetet kombinerar en rättsdogmatisk metod med två tvärvetenskapliga ram-verk: Bevisrättslig sannolikhetsteori (bayesiansk metod) samt rättsekonomi (cost-benefit-analys). Genom begreppet robusthet utreds huruvida en utred-ning kan anses tillräcklig för en fällande dom när det föreligger en generell eller specifik risk... (More)
Detta arbete undersöker hur den snabba utvecklingen av artificiell intelligens, med särskilt fokus på genererade bilder, påverkar bevisvärderingen och utred-ningsansvaret i svenska brottmål. Med utgångspunkt i principerna om fri be-visföring och fri bevisvärdering i 35 kap. 1 § rättegångsbalken analyseras hur domstolarna hanterar, eller bör hantera, invändningar från den tilltalade om att digital bevisning har manipulerats.
Arbetet kombinerar en rättsdogmatisk metod med två tvärvetenskapliga ram-verk: Bevisrättslig sannolikhetsteori (bayesiansk metod) samt rättsekonomi (cost-benefit-analys). Genom begreppet robusthet utreds huruvida en utred-ning kan anses tillräcklig för en fällande dom när det föreligger en generell eller specifik risk för digital manipulation. En utrednings robusthet brister om ytterligare hypotetiska utredningsåtgärder inte osannolikt skulle kunna för-ändra det befintliga bevisläget till den tilltalades fördel.
Uppsatsen applicerar dessa teoretiska modeller på svensk underrättspraxis. Analysen visar att en invändning om AI-manipulation inte sänker den sam-manlagda beviskraften om den övriga utredningen är tillräckligt robust. I såd-ana fall kan invändningen avfärdas som en efterhandskonstruktion. Om det digitala beviset däremot utgör den helt centrala bevisningen och utredningen i övrigt saknar stöd, blir en forensisk autentisering avgörande för att uppnå beviskravet.
Slutligen presenteras ett rättsekonomiskt riktmärke för åklagarens utrednings-ansvar under förundersökningen. Eftersom statens resurser är begränsade och en slentrianmässig autentisering av all digital bevisning skulle överbelasta NFC, bör forensiska äkthetsundersökningar begränsas utifrån en cost-benefit-bedömning. En undersökning är rättfärdigad endast när bevisets potential att förändra bevisläget (switch-ability) är reell, vilket avgörs genom att väga bevi-sets centrala roll mot hur dess digitala och fysiska hantering har dokumente-rats. (Less)
Please use this url to cite or link to this publication:
author
Wallanger, Jakob LU
supervisor
organization
alternative title
The impact of potentially AI-generated evidence on the standard of proof – A study on robustness and probability
course
JURM02 20261
year
type
H3 - Professional qualifications (4 Years - )
subject
keywords
Bevisvärdering, Processrätt, Straffrätt, Robusthet, AI
language
Swedish
id
9227201
date added to LUP
2026-06-19 09:55:03
date last changed
2026-06-19 09:55:03
@misc{9227201,
  abstract     = {{This thesis examines how the rapid advancement of artificial intelligence, with a specific focus on generated imagery, impacts evidence and the duty to inve-stigate within Swedish criminal procedure. Grounded in the principles of free admissibility of evidence and the free evaluation of evidence under Chapter 35, Section 1 of the Swedish Code of Judicial Procedure, the study analyzes how courts evaluate, or ought to evaluate, defense claims that digital evidence has been manipulated.
The study combines a legal dogmatic method with two interdisciplinary frameworks: Probabilistic evidence theory (the Bayesian method) and law and economics (cost-benefit analysis) Through the concept of robustness, the the-sis investigates whether an investigation can be deemed sufficient to support a conviction when facing a general or specific risk of digital manipulation. An investigation lacks robustness if additional, hypothetical investigation measu-res could reasonably alter the strength of the existing evidence in favor of the defendant. 
The thesis applies these theoretical models to contemporary Swedish lower court case law. The analyzis demonstrates that an objection regarding AI ma-nipulation does not diminish the aggregate probative value if the remaining investigation is sufficiently robust. In such scenarios, the objection can be dismissed as an ex post fabrication. Conversely, if the digital evidence i central to the prosecution’s case and the remaining investigation lacks independent corroboration, forensic authentication becomes essential to satisfy the standard of proof.
Lastly, the thesis establishes an economic benchmark for the prosecutor’s duty to investigate during the preliminary inquiry. Given that state resources are finite and a routine authentication of all digital evidence would overwhelm the National Forensic Centre, forensic verifications must be constrained by cost-benefit parameters. An investigation is justified only when the potential of the verification to alter the outcome of the case (switch-ability) is real, which is determined by balancing the centrality of the evidence against how its digital and physical handling has been documented.}},
  author       = {{Wallanger, Jakob}},
  language     = {{swe}},
  note         = {{Student Paper}},
  title        = {{Hur potentiellt AI-genererad bevisning påverkar beviskravet – En studie om robusthet och sannolikhet}},
  year         = {{2026}},
}