Expert insights on AI code detection and academic integrity
As AI coding assistants become ubiquitous, learn how institutions are adapting to detect AI-generated code and maintain educational standards.
Codequiry Editorial Team
·
Jan 5, 2026
Stay ahead with expert analysis and practical guides
General
9 min
Priya Sharma
·
9 hours ago
Three decades of code similarity detection can be told as two changes in the unit of comparison, from lines to token sequences to syntax trees. Each shift shrank the review queue by an order of magnitude. Here's what each era bought, what it still misses, and how generated code forced the field to add a second axis.
General
11 min
James Okafor
·
2 days ago
Renaming variables and extracting helpers is the cheapest way to defeat a naive similarity checker, and the fix is not a better diff. This walks through the normalization stack that makes refactoring-resistant plagiarism detection work, from token streams to winnowing to tree edit distance, and is honest about the precision you pay for each step.
General
10 min
James Okafor
·
6 days ago
An AI code detector is a measurement instrument, and like any instrument it needs a baseline before you trust the number it prints. This walks through per-file scoring, empirical false positive rates on pre-2023 submissions, cohort normalization, and why the AI score should never travel alone.
General
12 min
Priya Sharma
·
1 week ago
A working build guide to the similarity pipeline behind most academic code checks: tokenization, k-gram fingerprinting, winnowing, and AST shape comparison in roughly 300 lines of Python. Includes the threshold calibration table we used across six assignments and an honest account of where a homegrown checker stops being enough.
General
10 min
Priya Sharma
·
1 week ago
A take-home pipeline with 1,180 submissions and no AI screening is a coin flip dressed up as a hiring process. Here's what the scores actually measure, how to combine them with similarity checks, and where the whole method quietly falls apart.
General
15 min
Marcus Rodriguez
·
1 week ago
A code similarity score is not a probability of cheating. It's a measurement of overlapping token fingerprints, and it can be high for honest reasons or low for dishonest ones. Here's what sits underneath the number, with worked Java examples and the thresholds that hold up when a student appeals.
General
10 min
Alex Petrov
·
2 weeks ago
Pasting from Stack Overflow is the most common form of web code plagiarism, and peer comparison tools like MOSS will never see it. Here's how token fingerprinting actually matches a submission to a 2013 answer, how to read a web match report without overreacting, and why "Stack Overflow is public domain" is the most expensive misconception in the room.
General
12 min
Marcus Rodriguez
·
2 weeks ago
A 71% similarity score means nothing on its own, but it has to survive a room full of people who want an answer. Here is the exact calibration workflow I use: normalization settings, cohort percentiles, tiered thresholds, and an evidence packet that holds up when a student appeals.
General
13 min
Rachel Foster
·
3 weeks ago
License detection spent twenty years matching strings against license texts and missing the code underneath them. The algorithms that fixed it came out of plagiarism research, not legal tech. A look at what changed after 2003, and what the AI boom reopened.
General
10 min
Marcus Rodriguez
·
3 weeks ago
Meridian, a 40-person benefits SaaS, had 340,000 lines of contracted code and no way to answer question 34 on a diligence questionnaire. Here's the five-gate provenance audit that followed, what the tools actually found, and the clause in the MSA that would have made all of it cheap.
General
10 min
Alex Petrov
·
3 weeks ago
I ran 4,000 bootcamp submissions through web-source matching and 22% matched something public. Most of those matches were harmless. Here's what a real Stack Overflow diff looks like, why GitHub matches almost always come with a receipt, and how I stopped flagging half my class with false positives.
General
12 min
Alex Petrov
·
4 weeks ago
I seeded 48 known-copy pairs into 1,247 real bootcamp submissions and ran them through MOSS, JPlag, Dolos, and Codequiry. Token-based engines held up against renaming and reformatting, then collapsed once control flow changed. Here's the data, the methodology, and where every engine breaks.