papers/hanzo-fhe-inference/sections/09-security-analysis.tex

\section{Security Analysis}

\label{sec:security}
% ============================================================================

\begin{theorem}[IND-CPA Security]
The Hanzo FHE Inference system achieves IND-CPA security under the Ring
Learning with Errors (RLWE) assumption with parameters providing $\geq$128-bit
classical security as estimated by the LWE Estimator~\cite{lweestimator}.
\end{theorem}

\begin{remark}
FHE provides stronger guarantees than TEE-based approaches: security relies on
mathematical hardness assumptions (RLWE), not on hardware trust. The RLWE
problem is believed to be hard even for quantum adversaries, providing
post-quantum security~\cite{hanzopq}.
\end{remark}

% ============================================================================