AES-GCM Implementation Considerations
You'll learn: The critical implementation pitfalls that can completely break AES-GCM's security, even when used correctly at the API level.
The Three Deadly Implementation Sins
Even though AES-GCM is theoretically secure (as you learned in lesson 125), real-world implementations can fail catastrophically if you violate certain rules.
Nonce Reuse: The Fatal Mistake
Nonce reuse in AES-GCM is absolutely catastrophic. Unlike CBC mode where nonce reuse "only" leaks plaintext patterns, reusing a nonce in GCM allows attackers to:
- Recover the authentication key
- Forge valid authentication tags for arbitrary messages
- Decrypt all messages encrypted with that nonce
Think of the nonce as a lock combination — once an attacker sees the same combination twice, they can reverse-engineer the entire system. Never reuse a nonce with the same key, ever.
Tag Truncation: Weakening the Shield
AES-GCM produces a 128-bit authentication tag by default, but some implementations allow truncation to 96, 64, or even 32 bits to save bandwidth. Each bit you remove halves the security: a 64-bit tag gives an attacker a 1-in-2⁶⁴ chance of forging a valid message — feasible with enough attempts.
Best practice: Use the full 128-bit tag unless you have extraordinary constraints and understand the risk.
Constant-Time Verification: Preventing Timing Leaks
When verifying the authentication tag, you must compare all bytes in constant time. A naive byte-by-byte comparison that exits early on mismatch leaks information through timing, potentially allowing attackers to forge valid tags through millions of carefully timed attempts.
Key Takeaway: AES-GCM's security depends entirely on never reusing nonces, using full-length tags, and implementing constant-time tag verification — get any of these wrong and encryption becomes worthless.