1Qompass, Spokane, WA
2Arnot Ogden Medical Center Emergency Medicine Residency Program, Elmira, NY
3The Johns Hopkins University School of Medicine, Department of Orthopaedic Surgery, Baltimore, MD
▶️ 2025 American Medical Association (AMA) Challenge Poster & Writeup
The catastrophic Crowdstrike patch failure of July 19, 2024, exposed critical vulnerabilities in global healthcare systems, stemming from a memory safety issue in C++ code. This null pointer error, a common pitfall in languages without automatic memory management, led to system-wide failures in Microsoft-based environments while Linux/GNU and Apple systems remained unaffected. This event underscores the urgent need for robust, quantum-resistant cryptographic solutions in healthcare IT infrastructure.
We developed a protocol for building and benchmarking National Institute of Standards and Technology (NIST)-endorsed classical and post-quantum encryption algorithms on-premesis, using consumer grade Linux computers to prioritize viability for underserved regions & underfunded institutions. We compiled OpenSSL with Open Quantum Safe (OQS) C library to enable post-quantum encryption development that allowed the same level of access as Crowdstrike's faulty driver code while allowing for bindings with numerous memory safe programming languages. Our focus on post-quantum Key Encapsulation Mechanism (KEM) encryption reflects the ubiqutious protection that these protocols provide to secure communication and knowledge-work as well as the relative ease of hybridization with classical encryption protocols like Elliptical Curve Diffie-Hellman (ECDH). Following on-device compilation and installation of the encryption binaries, we built and executed an evaluation script with OpenSSL's native toolkit for twenty-four NIST-endorsed KEM protocols consisting of classical, quantum, and hybrid KEM implementations. We evaluated the KEMs on the number and rate of key generations (keygen), key encapsulation (encap) rate, and key decapsulations (decap) and rated their NIST post-quantum security level according to NIST advanced encryption standard (AES) exaustic key search levels.
We successfully benchmarked all 24 KEM protocols, producing an example public/private key pair following the evaluation. The 24 KEM protocols are evenly split across NIST security levels 1, 3, and 5, with 8 protocols at each.We made all relevant code, regulatory information, and the example cryptographic key pairs available on the Qompass AI Github page. We released them under the GNU Affero General Public License (AGPL) to maintain the free availability of these encryption tools to benefit communities.
| Algorithm | Type | NIST Security Level | Keygen (ms) | Encaps (ms) | Decaps (ms) | Keygens/s | Encaps/s | Decaps/s | Industry/Healthcare Usage |
|---|---|---|---|---|---|---|---|---|---|
| Frodo640AES | Quantum | Level 1 | 0.361 | 0.503 | 0.481 | 2773.0 | 1988.9 | 2081.0 | Experimental in IoT |
| Frodo640SHAKE | Quantum | Level 1 | 2.240 | 2.364 | 2.346 | 446.5 | 423.0 | 426.3 | Research in secure messaging |
| Frodo976AES | Quantum | Level 3 | 0.802 | 1.024 | 1.038 | 1247.0 | 976.8 | 963.6 | Tested in satellite communications |
| Frodo976SHAKE | Quantum | Level 3 | 4.975 | 5.208 | 5.128 | 201.0 | 192.0 | 195.0 | Evaluated for financial services |
| Frodo1344AES | Quantum | Level 5 | 1.350 | 1.656 | 1.599 | 741.0 | 604.0 | 625.3 | Considered for long-term data protection |
| Frodo1344SHAKE | Quantum | Level 5 | 8.772 | 9.174 | 9.009 | 114.0 | 109.0 | 111.0 | Evaluated for government communications |
| Kyber512 | Quantum | Level 1 | 0.022 | 0.021 | 0.017 | 44556.1 | 47830.3 | 58718.0 | Implemented in VPN services |
| Kyber768 | Quantum | Level 3 | 0.033 | 0.032 | 0.028 | 30291.8 | 31060.6 | 36305.1 | Tested in banking systems |
| Kyber1024 | Quantum | Level 5 | 0.045 | 0.045 | 0.040 | 22293.9 | 22075.8 | 24937.0 | Evaluated for aerospace industry |
| MLKEM512 | Quantum | Level 1 | 0.022 | 0.017 | 0.017 | 45416.2 | 59462.2 | 57611.1 | Research in smart home devices |
| MLKEM768 | Quantum | Level 3 | 0.036 | 0.027 | 0.027 | 28046.4 | 36703.0 | 37677.8 | Evaluated for telemedicine platforms |
| MLKEM1024 | Quantum | Level 5 | 0.045 | 0.040 | 0.042 | 22468.7 | 24869.7 | 23599.0 | Considered for national defense networks |
| BIKE-L1 | Quantum | Level 1 | 0.219 | 0.045 | 0.733 | 4556.0 | 22061.6 | 1364.6 | Experimental in IoT networks |
| BIKE-L3 | Quantum | Level 3 | 0.631 | 0.107 | 2.404 | 1586.0 | 9351.5 | 416.0 | Research in industrial control systems |
| BIKE-L5 | Quantum | Level 5 | 1.658 | 0.243 | 5.657 | 603.0 | 4123.0 | 176.8 | Evaluated for long-term data archiving |
| HQC-128 | Quantum | Level 1 | 1.828 | 3.613 | 5.882 | 547.0 | 276.8 | 170.0 | Research in wearable tech security |
| HQC-192 | Quantum | Level 3 | 5.525 | 10.989 | 16.949 | 181.0 | 91.0 | 59.0 | Evaluated for healthcare data exchange |
| HQC-256 | Quantum | Level 5 | 10.000 | 21.277 | 31.250 | 100.0 | 47.0 | 32.0 | Considered for military communications |
| P-256 + Kyber512 | Hybrid | Level 1 | 0.771 | 0.162 | 0.376 | 1296.9 | 6183.8 | 2658.2 | Tested in e-commerce platforms |
| P-384 + Kyber768 | Hybrid | Level 3 | 1.070 | 1.071 | 1.119 | 934.3 | 933.3 | 894.0 | Evaluated for cloud storage services |
| P-521 + Kyber1024 | Hybrid | Level 5 | 0.959 | 0.991 | 1.061 | 1042.9 | 1009.1 | 942.4 | Research in quantum-resistant blockchains |
| X25519 + Kyber512 | Hybrid | Level 1 | 0.071 | 0.105 | 0.099 | 14135.7 | 9543.4 | 10106.1 | Implemented in secure messaging apps |
| X25519 + Kyber768 | Hybrid | Level 3 | 0.086 | 0.115 | 0.111 | 11621.4 | 8704.0 | 9040.0 | Evaluated for VPN services |
| X448 + Kyber768 | Hybrid | Level 3 | 0.274 | 0.487 | 0.491 | 3644.9 | 2055.1 | 2037.4 | Research in high-security financial systems |
| Term | Explanation | Security Implication |
|---|---|---|
| Algorithm | The name of the encryption method used to secure data | N/A |
| NIST Security Level | Indicates the level of security as defined by NIST | Higher is more secure |
| Keygen (ms) | Time taken to generate a key pair (in milliseconds) | Lower is generally better, but too low may indicate weakness |
| Encaps (ms) | Time taken to encapsulate (encrypt) a shared secret (in milliseconds) | Lower is better for performance, but should balance with security |
| Decaps (ms) | Time taken to decapsulate (decrypt) a shared secret (in milliseconds) | Lower is better for performance, but should balance with security |
| Keygens/s | Number of key pairs that can be generated per second | Higher is better for performance, but should balance with security |
| Encaps/s | Number of encapsulations that can be performed per second | Higher is better for performance, but should balance with security |
| Decaps/s | Number of decapsulations that can be performed per second | Higher is better for performance, but should balance with security |
| Industry/Healthcare Usage | Examples of current or potential use in industry or healthcare | N/A |
| Level | Description | Healthcare Example |
|---|---|---|
| Level 1 | At least as hard to break as AES-128 | Securing patient portals |
| Level 3 | At least as hard to break as AES-192 | Protecting electronic health records (EHRs) |
| Level 5 | At least as hard to break as AES-256 | Safeguarding genomic data |
Note: While higher security levels provide stronger protection, they often come with increased computational costs. The choice of security level should be based on the sensitivity of the data and the specific requirements of the healthcare application.
Out of the evaluated KEMs, we propose hybrid combinations of ECDH and Kyber for most acute adoption of enhanced encryption protocols due to the layered security of nascent post-quantum encryption with established efficient classical protocols. Currently, Google Chrome implements X25519_Kyber768 hybrid encryption as part of its Transport Layer Security (TLS), offering a familiar and accessible platform to perform institutional assessements.
- Password authenticated key exchange-based on Kyber for mobile devices
- Post-quantum healthcare: A roadmap for cybersecurity resilience in medical data
- Transitioning organizations to post-quantum cryptography
We would like to thank the Ruth Jackson Orthopaedic Society and Zimmer Biomet for their generous support of our work.
Frequently Asked Questions
Q: How do you mitigate against bias?
TLDR - we do math to make AI ethically useful
A: We delineate between mathematical bias (MB) - a fundamental parameter in neural network equations - and algorithmic/social bias (ASB). While MB is optimized during model training through backpropagation, ASB requires careful consideration of data sources, model architecture, and deployment strategies. We implement attention mechanisms for improved input processing and use legal open-source data and secure web-search APIs to help mitigate ASB.
AAMC AI Guidelines | One way to align AI against ASB
AI Math at a glance
Forward Propagation Algorithm
$$ y = w_1x_1 + w_2x_2 + ... + w_nx_n + b $$
Where:
-
$y$ represents the model output -
$(x_1, x_2, ..., x_n)$ are input features -
$(w_1, w_2, ..., w_n)$ are feature weights -
$b$ is the bias term
Neural Network Activation
For neural networks, the bias term is incorporated before activation:
$$ z = \sum_{i=1}^{n} w_ix_i + b $$
$$ a = \sigma(z) $$
Where:
-
$z$ is the weighted sum plus bias -
$a$ is the activation output -
$\sigma$ is the activation function
Attention Mechanism- aka what makes the Transformer (The "T" in ChatGPT) powerful
The Attention mechanism equation is:
$$ \text{Attention}(Q, K, V) = \text{softmax}\left( \frac{QK^T}{\sqrt{d_k}} \right) V $$
Where:
-
$Q$ represents the Query matrix -
$K$ represents the Key matrix -
$V$ represents the Value matrix -
$d_k$ is the dimension of the key vectors -
$\text{softmax}(\cdot)$ normalizes scores to sum to 1
Q: Do I have to buy a Linux computer to use this? I don't have time for that!
A: No. You can run Linux and/or the tools we share alongside your existing operating system:
- Windows users can use Windows Subsystem for Linux WSL
- Mac users can use Homebrew
- The code-base instructions were developed with both beginners and advanced users in mind.
Q: Do you have to get a masters in AI?
A: Not if you don't want to. To get competent enough to get past ChatGPT dependence at least, you just need a computer and a beginning's mindset. Huggingface is a good place to start.
Q: What makes a "small" AI model?
A: AI models ~=10 billion(10B) parameters and below. For comparison, OpenAI's GPT4o contains approximately 200B parameters.
This project is licensed under the Apache License, Version 2.0.