Skip to content

Research

Criminal law was built around the human actor. Intelligent systems are testing that foundation.

My research examines how criminal law should respond when agency, evidence and responsibility are transformed by artificial intelligence, robotics and digital infrastructures.

The central question is not only whether existing criminal-law categories can be applied to new technologies. It is whether those categories still describe the structure of action, control and blame in systems where conduct is distributed across humans, machines, developers, operators, institutions and data environments.

Janko Munjić speaking on artificial intelligence and criminal law
Speaking on AI and criminal law

Research architecture

From technologically mediated action to the safeguards that constrain its legal treatment.

Five analytical stages organise six connected strands of work. The order is a way of reading the programme, not a sequence each project follows: the strands overlap, and several works belong to more than one.

Research Themes

Six strands of work, from attribution to procedural safeguards.

Current Projects

Work in progress.

In development

The Limits of Humancentrism in Criminal Law

Interrogating the assumption that every punishable act can be traced to a bounded and identifiable human agent in a posthuman environment.

Submitted Research Proposals

Whether AI changes how lawyers reason — and whether the change outlasts the tool.

SOVEREIGN LEGAL AI

Legal Reasoning Sovereignty and Jurisdictional Drift in the Age of Large Language Models

ERC Plus 2026 proposal — submitted, under evaluation

No funding has been awarded and no evaluation result has been received.

A legal system can change while its formal rules stay still. General-purpose AI can reach a locally defensible answer along a reasoning route shaped by another jurisdiction, and the proposed seven-year programme would test whether that route enters professional legal reasoning, persists after direct model use ends, and circulates through later human work — holding legal quality separate, since a model may improve accuracy and still change which sources are chosen and how doctrine is framed. Serbia would provide the deep pilot, Italy a dense EU civil-law comparator whose law reserves core judicial interpretation and the assessment of fact and evidence to the human magistrate, and Montenegro a separate legal order in close linguistic proximity to Serbia; criminal law and procedure supply the high-constraint domain, private and commercial law the structural comparison. It carries contestability, AI-mediated reasoning and the transformation of legal judgment — the strands above — into a comparative, empirical design.

Role
Principal Investigator
Call
ERC-2026-PLUS
Host Institution
Institute of Comparative Law, Belgrade
Duration proposed
84 months
Submitted
1 September 2026
Panels
SH2 Institutions, Governance and Legal Systems (primary); PE6 Computer Science and Informatics (secondary)

From Shadow AI to European Legal Capability: Building a Jurisdictional Drift Audit for General-Purpose AI

Charlemagne Prize Academy Fellowship 2026/27 proposal — submitted, under evaluation

No funding has been awarded and no evaluation result has been received.

A smaller precursor programme: developing the Jurisdictional Drift Audit itself and running a limited comparative validation. It is the first step of the same agenda — SOVEREIGN LEGAL AI is what adds the seven-year causal design, delayed persistence, professional transmission and three-jurisdiction replication.

Role
Individual applicant, proposed Fellow
Duration proposed
12 months
Period
November 2026 – October 2027, if awarded
Submitted
15 July 2026

Selected Questions

The questions driving the work.

  • When conduct is distributed across humans, machines, developers and operators, who is the criminal-law subject?

  • Can intention and control be meaningfully assessed in neuroprosthetic and augmented-agency systems?

  • How can adversarial review survive encrypted-platform and algorithmically processed evidence?

  • What court-usable tests can attribute responsibility without abandoning legality and culpability?

  • How should human-rights safeguards constrain AI-enabled judicial cooperation?

Methodological Angle

Court-usable, not only speculative — frameworks tested against procedural reality, the limits of institutions and the demands of legality and human rights.

See the publications