evidoria

← Back to browse

Good practice

Poland's Uniform Anti-Plagiarism System (JSA) — National AI-Text Detection Infrastructure for University Theses

Poland · Warsaw · See the Poland profile

Since 2019 Poland has required every university to screen theses through JSA, a free national anti-plagiarism system used by 400+ institutions and 100,000+ users. A Feb 2024 update added AI-text detection, though officials stress it cannot replace supervisor judgement.

1.5 million
Theses examined (2019–2023 (four years))
320,000+
Theses flagged for potential plagiarism (2019–2023 (four years))
400+
Institutions using JSA
100,000+
Registered users
10 billion+
Microdocuments in comparison corpus
3 million+
Works in National Repository of Diploma Theses
~900 million
Polish web documents in NEKST database
11 PLN million
Funding for JSA build (2017–2019)
~1 in 5
Students planning to use AI tools while writing thesis
68 %
Students intending to use ChatGPT-style tools during studies generally

Details

Maturity
Established
Promoter
National Information Processing Institute (OPI-PIB) / Polish Ministry of Science and Higher Education
Period
2017–ongoing (mandatory nationwide since Jan 2019; AI-detection module added Feb 2024)
Keywords
plagiarism detection, AI-text detection, higher education policy, academic integrity, national infrastructure

Context

Poland's Higher Education Law has required every university and doctoral school to screen theses through a single national system since January 2019. JSA was built in 2017–2019 by the National Information Processing Institute (OPI-PIB) under Dr Marek Kozłowski, funded with PLN 11 million from an EU-backed operational programme.

Objectives

To provide a free, nationally mandated plagiarism-screening infrastructure covering every thesis submitted at Polish universities, and — from February 2024 — to add detection of AI-generated text in response to rising student use of AI writing tools.

Activities

JSA compares submitted texts against a corpus of more than 10 billion "microdocuments" drawn from the National Repository of Diploma Theses (3+ million works), the NEKST database of roughly 900 million Polish web documents, six Wikipedia language editions, legal-act databases and open-access journals, breaking texts into disordered word collections to resist simple paraphrasing. In February 2024, OPI added a module to flag AI-generated text, prompted partly by a Digital Care survey finding about one in five students planned to use AI tools while writing their thesis.

Results

In four years of operation, JSA examined about 1.5 million theses across 400+ institutions and over 100,000 registered users, flagging more than 320,000 as containing potential plagiarism — a sharp jump in coverage from the roughly 30–40% of theses that received any plagiarism screening before JSA became mandatory.

Conclusions

JSA's documented strength is administrative and institutional — nationwide coverage, scale and legal mandate — rather than a measured pedagogical or learning-outcome effect. Officials are candid about the AI-detection module's limits: deputy director Marek Michajłowicz noted it's unlikely that AI-generated text can be reliably recognised, so the system supports rather than replaces supervisor judgement. No independent, peer-reviewed validation of the AI-detection module's false-positive or false-negative rate has been published.

Implementation
Indicative cost
Medium (€50k–€500k) — PLN 11 million (EU-backed Knowledge Education Development Operational Programme) for the 2017–2019 build; ongoing operating/maintenance costs and AI-detection module development not separately disclosed.
Time to results
Long (> 3 years) — Built 2017–2019, mandatory nationwide from January 2019, AI-text detection module added February 2024; ongoing.
Staffing & skills
National Information Processing Institute (OPI-PIB) technical team led by Dr Marek Kozłowski, University thesis supervisors and administrators using the system

Conditions for success

  • Legal mandate (Higher Education Law) requiring universal use
  • Centralised national corpus (theses repository, web database, Wikipedia, legal acts, journals) for comparison
  • Sustained public funding for corpus maintenance and updates (e.g., AI-detection module)

Common failure modes

  • AI-generated text is not reliably detectable by the module or the human eye, per OPI's own deputy director
  • No independent validation of false-positive/false-negative rates published

Where it fits

Governance type
national government / public research institute
Scale
national
Income level
high-income

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful