Update README.md

README.md

@@ -72,6 +72,16 @@ While titled **RobustMNIST**, it is important to clarify that "robust" does not
 - **Semantic Edge Cases:** Certain transformations, such as rotating a "6" until it looks like a "9" or mirroring asymmetric digits create mathematical ambiguities. We acknowledge these limits; at this parameter count, the model prioritizes identifying structured digits over handling every possible topological distortion.
 - **Research Scope:** This is a 1.0 release focused on balancing clean accuracy with OOD calibration. We agree that edge cases exist where the model may still fail or default to "Unknown" unexpectedly.
 
+## CRUCIAL SAFETY WARNING, DISCLAIMERS, & LIABILITY LIMITATION
+
+This model is an experimental scientific artifact provided strictly "as is" and "with all faults." The developers, distributors, and contributors disclaim all warranties, express or implied, including but not limited to any implied warranties of merchantability, fitness for a particular purpose, or non-infringement.
+
+*   **NOT FOR USE IN SAFETY-CRITICAL OR MISSION-CRITICAL SYSTEMS:** Under no circumstances should this model be deployed in systems where failure could lead to physical injury, loss of life, property damage, or financial loss. This includes, but is not limited to: autonomous driving systems, industrial robotics, automated sorting, optical character recognition (OCR) for medical prescriptions, security and surveillance monitoring, biometric authentication, or hazardous materials handling.
+*   **THE "ROBUSTNESS" ILLUSION & ADVERSARIAL VULNERABILITY:** Users must not equate the comparative term "Robust" with mathematical invulnerability. Due to its extremely small parameter scale (approximately 430k parameters), this model is highly vulnerable to targeted adversarial perturbations (such as Fast Gradient Sign Method, Projected Gradient Descent, or single-pixel modifications). Imperceptible image noise can easily fool the model into misclassifying numbers or silently bypassing Class 10 entirely.
+*   **FALSE SENSE OF SECURITY FROM "CLASS 10" (THE UNKNOWN CLASS):** The inclusion of an "Unknown" class does not establish a reliable safety barrier. There are infinite possible out-of-distribution structures, backgrounds, geometric patterns, or optical anomalies that will completely evade Class 10 detection and register as high-confidence false-positive predictions within Classes 0–9.
+*   **VULNERABILITY TO GEOMETRIC AND LIGHTING DISTORTIONS:** The model lacks rotation invariance and topological orientation awareness. Rotations, severe shearing, scaling anomalies, and hardware camera sensor artifacts will result in silent classification failures without raising exceptions or warnings.
+*   **COMPLETE EXCLUSION OF LIABILITY:** By downloading, executing, or incorporating this model or its weights, the end-user assumes 100% of all legal, operational, and financial risks. The creators and distributors shall not be held liable for any direct, indirect, incidental, special, or consequential damages (including but not limited to hardware failure, operational disruptions, algorithmic bias propagation, or security breaches) arising from the use of this model.
+
 ## Usage
 
 To use this model, ensure you have `model.py` and `model.pt` in your directory.