.png){kind=logo}
AI Detection
/ Python Algorithm Implementation - AI Detection Case Study
How to read these scores
Model v3e, updated Sep 24, 2026
An AI score is a signal, not a verdict: a 0 to 100% probability from stylistic patterns common to ChatGPT, Claude & Copilot. Starter code, very clean code, and short files can read high, so treat a high score as a reason to open the submission and start a conversation, not as proof.
Open a flagged submission and read the highlighted regions before acting. Those lines are what the score is resting on. If they cover only imports, licence headers or boilerplate, the score is weak evidence however high it reads.
How accurate is this?
90%
AI/GPT-written code caught.
Trained on Claude Opus 5.5, GPT-5.6 Sol and GPT-5.5 Codex submissions.
1.3%
Human-written code falsely flagged.
Low, not zero, so read the highlighted lines before acting on a score.
Detection can be weaker against other models, and we cannot predict in advance which human files will read high.
39%
Average AI Score
8 submissions analyzed
8
Submissions
99
Files
85%
Highest File Score
1
Flagged
Distribution by AI score
8 analyzed
Low Risk 0 to 39%
6
Medium 40 to 69%
1
High Risk 70%+
1
Submissions
| Submission | AI Score | Range | Files | Status | |
|---|---|---|---|---|---|
|
12_bertil291utn_bubble-sort-proj...
10 months ago
|
28%
|
11 to 42% | 5 | Safe | |
|
10_Justintime50_algorithms
10 months ago
|
42%
|
16 to 74% | 26 | Moderate | |
|
12_Nikoo-Asadnejad_SortingAlgori...
10 months ago
|
34%
|
8 to 50% | 35 | Safe | |
|
11_SlowJii_Workshop-FPT
10 months ago
|
18%
|
7 to 45% | 5 | Safe | |
|
11_AbdallahHemdan_Sorting-Visual...
10 months ago
|
33%
|
10 to 65% | 6 | Safe | |
|
13_addyosmani_bubblesort
10 months ago
|
36%
|
19 to 50% | 11 | Safe | |
|
13_jefelewis_algorithms-review
10 months ago
|
39%
|
14 to 55% | 10 | Safe | |
|
javabuble-written-by-gpt5
10 months ago
|
85%
|
85% | 1 | Review |
Submission Details
Loading analysis...