manta¶
Weight steganography research platform. A workspace for studying covert information channels in neural network weights and the defenses against them.
Public site is intentionally minimal
Detailed methodology, empirical results, operator-facing tradecraft, and source code are part of an in-progress academic submission and a private research repository. They are withheld from this public site until the corresponding paper is publicly released.
For coordinated-disclosure inquiries, contact the project maintainers through the repository.
Scope¶
manta is an academic research codebase focused on:
- Characterising the steganographic capacity of neural network weights.
- Building and evaluating detectors for weight-level covert channels.
- Studying the supply-chain implications for distributed model artifacts.
The work is conducted with explicit defensive intent. All offensive material is gated behind embargo, with a focus on giving defenders detection capability rather than enabling abuse.
Status¶
Active research. Public materials will expand once the associated paper clears peer review.