The engine that
sees through obfuscation.
Zeus Hyper uses machine learning to understand code semantically—not just textually. Rename variables, shuffle functions, translate between languages. It doesn't matter. Zeus Hyper sees the logic underneath.
# Student renamed everything
# Zeus Hyper still finds the match
def calculate(x, y):
result = x * y + 42
return result
# → 94% match to GitHub repoNot pattern matching. Understanding.
ML-Based Tokenization
Zeus Hyper converts code into semantic tokens that represent logic, not syntax. Variable names become irrelevant. The algorithm matches meaning.
Cross-Language Detection
Python to Java. C++ to Rust. Zeus Hyper understands the underlying algorithm regardless of which language it's written in.
Parallel Processing
Multi-threaded Node.js engine processes thousands of files simultaneously. A 500-student class takes seconds, not hours.
How Zeus Hyper processes code
Four stages transform raw source into comparable semantic fingerprints.
Parse
Language-specific parsers extract the AST. Comments and whitespace stripped. Pure structure remains.
Normalize
Variables become placeholders. Function names become generic tokens. The code is reduced to its logical skeleton.
Fingerprint
ML models generate semantic hashes. Similar logic produces similar fingerprints—regardless of surface changes.
Compare
Fingerprints are matched against peers, web sources, and known AI patterns. Matches ranked by confidence.
What Zeus Hyper catches
Every evasion technique students try. Zeus Hyper has seen it before.
Obfuscation Detection
Students try to hide. Zeus Hyper finds them anyway.
Variable renaming, function reordering, dead code insertion, loop restructuring—these tricks fool legacy tools. Zeus Hyper tokenizes at the semantic level, so surface changes don't affect detection.
- Variable/function renaming
- Code block reordering
- Dead code insertion
- Loop/conditional restructuring
ChatGPT & Copilot
Trained on millions of AI-generated samples
LLMs have telltale patterns—specific comment styles, naming conventions, code structures. Our ML models recognize these fingerprints with 95%+ accuracy.
- GPT-4 / GPT-4o detection
- Claude pattern matching
- GitHub Copilot signatures
- Confidence scoring
Zeus Hyper also includes access to MOSS, JPlag, and Dolos for institutions that want comparison baselines.
Zeus Hyper vs. legacy tools
"MOSS would take 3-4 days for our 400-student intro class. Zeus Hyper does it in under a minute. But the real win was catching students copying from GitHub and using ChatGPT—MOSS had no answer for that."
For the curious
What ML models does Zeus Hyper use?
Zeus Hyper uses a combination of transformer-based embeddings for semantic similarity and custom-trained classifiers for AI detection. The tokenization layer is proprietary but inspired by code2vec research.
How does cross-language detection work?
Code is normalized to cross-family tokens representing operations (assignment, loop, conditional, function call, etc.). Two implementations of the same algorithm—in Python and Java, for example—produce nearly identical token sequences.
What's the false positive rate?
Under 2% in our benchmarks. We tune for high precision because false accusations are worse than missed catches. Every match includes confidence scores so you can set your own thresholds.
Can students beat Zeus Hyper?
Not with known techniques. We've tested against every obfuscation method in academic literature plus novel approaches from red-team exercises. The semantic approach is fundamentally harder to fool than text matching.
How is my code stored?
Encrypted at rest (AES-256), encrypted in transit (TLS 1.3), auto-deleted after 90 days (configurable). FERPA compliant. We never use your code to train models.
Can I access Zeus Hyper via API?
Yes. Full REST API with batch processing support. See the docs →
See Zeus Hyper in action
Run your first check in under 2 minutes.