Post-Quantum Cryptography for AI Systems: Preparing for the Quantum Threat
Quantum computers capable of breaking RSA-2048 and ECC are estimated to arrive by 2030-2035. AI systems that transmit model weights, training data, and inference results over encrypted channels need to prepare now.
1. What's at Risk
| Asset | Current Protection | Quantum Threat |
|---|---|---|
| Model weights in transit | TLS 1.3 (ECDHE) | Broken by Shor's algorithm |
| API keys and tokens | RSA-2048 signatures | Broken by Shor's algorithm |
| Training data | AES-256 encryption | Weakened by Grover's (128-bit effective) |
| Stored encrypted models | RSA key wrapping | "Harvest now, decrypt later" attacks |
2. NIST Post-Quantum Standards
ML-KEM (Kyber) → Key encapsulation (replacing ECDH key exchange)
ML-DSA (Dilithium) → Digital signatures (replacing RSA/ECDSA signatures)
SLH-DSA (SPHINCS+) → Hash-based signatures (stateless fallback)
3. Migration Path
# Hybrid TLS configuration (classical + post-quantum)
tls:
min_version: "1.3"
key_exchange:
- "X25519Kyber768" # Hybrid: classical X25519 + post-quantum Kyber
signature:
- "ML-DSA-65" # Post-quantum signature
- "ECDSA-P256" # Classical fallback
The time to adopt post-quantum cryptography is now — not after quantum computers break your encryption.
