aes-gcm-siv · shipping · also AEAD_AES_128_GCM_SIV, AEAD_AES_256_GCM_SIV

AES-GCM-SIV

AES-GCM-SIV is an authenticated-encryption construction that limits the damage from accidental nonce reuse using a synthetic initialization vector.

RFC 8452 specifies 128- and 256-bit keys, a 96-bit nonce and a 128-bit tag. POLYVAL authenticates plaintext and associated data before counter-mode encryption. Message keys depend on the master key and nonce, so repeating the nonce repeats those derived keys. [1] [2]

With the same key, nonce and associated data, identical plaintext produces identical ciphertext. Misuse resistance does not promise unlimited reuse: the specification recommends random nonces, with security bounds depending on repetition, message count and length. [1]

In the linked native-library example, altered ciphertext and a changed expected version are rejected, but an old ciphertext still decrypts with its old context. Rejecting a rolled-back version requires an independently trusted freshness policy, not merely an authenticated version field stored beside the data. [4] [3]

Decryption computes candidate plaintext but must not release it until authentication succeeds. Ciphertext includes a 16-byte tag, so its length limit exceeds the plaintext limit. The example's 1 MiB payload cap is a resource limit, not a cryptographic usage bound. [1] [8]

Rejecting wrong keys in tests is not a proof of key commitment. That property resists a ciphertext valid under different keys. Nonce misuse resistance does not imply it: the 2021 result of Len et al. is untargeted and cannot precisely choose the colliding key set, not a recovery of arbitrary random secret keys. [9] [10] Albertini et al. describe a separate construction for two known keys. [11]

Relations

related
Cryptographic nonce, Authenticated encryption, Key commitment

Limitations

  • Encryption needs two passes over the message; it cannot emit ciphertext before processing the complete plaintext. [1]
  • Successful vector tests do not prove security bounds or establish a safe storage protocol. [8]

Citations

  1. RFC 8452: AES-GCM-SIV, sections 4-6 and 9, Appendices B-C (accessed 2026-11-21)
  2. Gueron and Lindell. Better Bounds for Block Cipher Modes of Operation via Nonce-Based Key Derivation. CCS 2017, section 6. (accessed 2026-11-21)
  3. RFC 5116: AEAD scope and the separation of anti-replay services, section 1.2 (accessed 2026-11-21)
  4. Hesela: native-library nonce, context and replay observations with public test inputs (accessed 2026-11-21)
  5. Hesela: reproduction instructions and limits of the demonstration (accessed 2026-11-21)
  6. RFC 9771: key commitment and its distinction from full commitment, section 4.3.3 (accessed 2026-11-21)
  7. Len, Grubbs and Ristenpart: Partitioning Oracle Attacks, USENIX Security 2021, section 3.2 (accessed 2026-11-21)
  8. Albertini et al.: How to Abuse and Fix Authenticated Encryption Without Key Commitment. USENIX Security 2022, section 3.4.1. (accessed 2026-11-22)

Further reading

  1. Hesela: bounded AES-GCM-SIV and AES-GCM demonstration using Node/OpenSSL (accessed 2026-11-21)
  2. Hesela: reference-vector, corruption, context and malformed-input tests (accessed 2026-11-21)
  3. RFC 8452 Appendix C: extracted numeric test vectors with source digest (accessed 2026-11-21)
Sources & caveats
  • RFC 8452 is an Informational CFRG specification, distinct from the earlier GCM-SIV paper and AES-SIV in RFC 5297. [1]
  • The CCS 2017 paper supplies a concrete cryptographic analysis, not a storage-fleet experiment. Its historical performance figures are not repeated as current benchmarks. [2]
  • The example runs established library primitives with synthetic public inputs; it is not a new cryptographic implementation or proof. [4]

JSON: /data/terms/aes-gcm-siv.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. AES-GCM-SIV. Hesela. 2026-11-22. https://hesela.dev/terms/aes-gcm-siv/