The startup develops a software development kit that secures on-device AI models from reverse engineering, protecting intellectual property across platforms such as smartphones and desktops. By preventing unauthorized access and duplication of AI models, the solution helps clients maintain the integrity and competitive advantage of their proprietary technologies.
Funding
$440K raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
AI models deployed on edge devices and servers are vulnerable to reverse engineering, allowing malicious actors to steal proprietary algorithms and intellectual property. Existing security measures like encryption are often insufficient, as models must be decrypted during inference, creating opportunities for runtime attacks and model extraction. This exposure leads to intellectual property theft, unauthorized reuse, and advanced attacks such as adversarial examples and model inversion.
Solution
Skyld provides a software development kit (SDK) that secures on-device AI models against reverse engineering, protecting intellectual property across various deployment environments. The SDK safeguards AI models both at rest and during execution, providing comprehensive protection against static and dynamic reverse-engineering techniques. By applying robust linear algebra transformations, Skyld prevents software analysis and AI attacks from accessing key model information, especially the weights. The SDK integrates into existing development pipelines with minimal disruption, turning trained models into protected models with only a few lines of code.
Target Audience
Skyld targets AI developers and enterprises deploying machine learning models on edge devices, IoT devices, smartphones, desktops, and on-premise servers who need to protect their intellectual property and maintain a competitive advantage.
Features
- Comprehensive protection against static and dynamic reverse-engineering
- Hardware agnostic, supporting deployment on smartphones, connected objects, desktops, web browsers, and on-premise servers
- Low-code integration, transforming trained models into protected models in less than 10 lines of code
- Compatibility with various operating systems, including Android, Linux, and Windows
- Support for machine learning inference frameworks such as ONNX, TensorFlow(lite), and Keras
- Protection for different kinds of neural networks, including CNN, RNN, LSTM, Transformers, Vision Transformers, and LLMs
- Negligible impact on model accuracy
- Performance overhead generally below 20%, depending on the specific model architecture