Reaction Attack on CKKS Deployments with Asymptotically Optimal Query Complexity
summary
The gist
The gist The attack recovers ternary secrets in query complexity that tightly matches theoretical lower bounds for CKKS deployments exposing decoded-slot validity responses and C2S rotation keys.
In short
The study shows that combining public coefficient-to-slot keys with a client's 1-bit validity response allows an attacker to recover ternary secret keys in asymptotically optimal query complexity. This attack exploits the interaction between validity checks and public homomorphic transformations, proving that observable reactions can leak secret information without breaking the underlying RLWE security.
Key concepts
- Reaction Attack
- This is a protocol-level side-channel where an attacker monitors how a system reacts (e.g., accept/reject) to specific inputs. In this case, the attacker uses these reactions, combined with public keys, to deduce parts of the secret key through repeated queries.
- Coefficient-to-Slot (C2S) Evaluation Keys
- These are public keys used in CKKS that map a coefficient in the ciphertext to a specific slot. The attack exploits how these keys interact with client validity responses, allowing an attacker to link guessed secret values to specific slots through these public transformations.
- Query Complexity Optimality
- The paper demonstrates that the number of queries needed for this key recovery attack is asymptotically optimal, meaning it matches the theoretical lower bounds for finding a secret. For uniform secrets, it uses O(N) queries, which is very close to the best possible performance achievable.
Terminology used across episodes
This episode discusses
The paper
Reaction Attack on CKKS Deployments with Asymptotically Optimal Query Complexity · Read on arXiv
Song Bian, Hyeongmin Choe, Jiseung Kim, Changmin Lee
Beihang University · University of Luxembourg · Jeonbuk National University · Korea University
Transcript
Introduction to the show: ident: Security Radio. Generated commentary on the latest security and cryptography papers.
Nadia: Today's paper: "Reaction Attack on CKKS Deployments with Asymptotically Optimal Query Complexity".
Elias: The gist The attack recovers ternary secrets in query complexity that tightly matches theoretical lower bounds for CKKS deployments exposing decoded-slot validity responses and C2S rotation keys.
Nadia: First, who's behind it and why it matters.
Paper summary: Elias: We've covered the details of the paper "Reaction Attack on CKKS Deployments with Asymptotically Optimal Query Complexity," focusing on how it exploits the interaction between C2S keys and validity responses <ref:2610.21127#pg1>.
Nadia: The authors are showing that this attack doesn't invalidate the security of RLWE or CKKS IND-CPA schemes, but rather that it reveals a specific protocol leakage surface <ref:2610.21127#pg3>.
Priya: So, if we take this for a moment, what does this imply for the real world deployment of these FHE schemes?
Elias: It implies that security analysis needs to look beyond the mathematical hardness of the underlying problem and focus on these protocol-level interactions between client checks and public transformations <ref:2610.21127#pg3>.
Nadia: The title itself, "Reaction Attack on CKKS Deployments with Asymptotically Optimal Query Complexity," really highlights that this isn't just some abstract theoretical attack, but something you can actually run against a deployed system with very efficient query counts <ref:2610.21127#pg3>.
Priya: So, the main conclusion is that observable validity reactions become a practical channel for recovering secret information in CKKS deployments <ref:2610.21127#pg3>.
Elias: The paper suggests that for defense, standardizing the externally visible response and monitoring those reactions per key is a practical way to mitigate this risk <ref:2610.21127#pg3>.
Conclusion: Nadia: So, we've seen how this reaction attack works on CKKS deployments using C2S keys and validity responses in this paper <ref:2610.21127#pg3>.
Elias: It really shows that you can exploit the way client checks interact with those public homomorphic linear transformations.
Priya: So, it's not breaking the core math of RLWE or CKKS IND-CPA security itself, but it leaks something about how those components are put together in a real deployment setup.
Nadia: Exactly. The paper suggests this is a protocol-level leakage surface that we need to look at <ref:2610.21127#pg3>.
Elias: They achieve what they call asymptotically optimal query complexity, which means the number of queries they need scales very well with the secret key entropy H.
Priya: So, for someone just listening, what does that mean? It means a specific kind of observable behavior in a deployed system can be used to recover secret information <ref:2610.21127#pg3>.
Nadia: It changes things because it proves that the client's validity reactions aren't just for checking if something is good, they can actually give away pieces of the secret key if you query them right <ref:2610.21127#pg3>.
Elias: The authors show that this attack doesn't need to reuse a fixed ciphertext; they use adaptive queries to find the support and then recover signs, which is pretty clever <ref:2610.21127#pg4>.
Priya: That optimization for sparse secrets, using a combinatorial group testing strategy to find the non-zero coefficients, that's interesting data showing how practical this attack is <ref:2610.21127#pg5>.
Nadia: It shows that observable validity reactions form a real key recovery channel in CKKS deployments even when the underlying math holds up <ref:2610.21127#pg3>.
Elias: And the defense they suggest—standardizing the visible response and monitoring those reactions per key—is what they say is the most practical way to stop it <ref:2610.21127#pg6>.
Priya: So, if you're a developer looking at deploying FHE, this paper tells you to watch your reaction rates closely <ref:2610.21127#pg6>.
Nadia: Right. And that sets us up for talking about how we can actually build those detection mechanisms next.
More episodes
- 2610.20952-Does Client Diversity Protect the Lightning Network? Evidence from the 2026 Core Lightning Shutdown
- 2610.20831-GhostLock: Learning-Assisted Blind GPS Signal Nullification Without Victim-Side Information
- 2610.20962-PAGE404: Breaking GPU Isolation Through Hardware-assisted Page Migration
- 2610.20907-From Detection to Decision: Calibration and Deferral for Prompt-Injection Detection
- 2610.20827-Performance Characterization of Pulsar's Software-defined Watermark Verifier
- 2610.21216-On the Complexity of Symmetric Boolean Functions in Multiparty Private Simultaneous Messages
- 2610.22047-SMT-Boosted Security Types and Correctness for Low-Level MPC
- 2610.21178-Explicit reduced APN permutations: ramification and optimal Galois-ring lifts
- 2610.21458-What Survives Self-Evolution: Provenance Marks in Copied Agent Harnesses
- 2610.21309-Combining LLM and Dynamic Analysis for Universal Testing of Static Application Security Testing Tools