mirror of
https://github.com/lux-apps/papers.git
synced 2026-10-30 18:15:21 +00:00
zeekay
abf99d05fc
Restructured from flat .tex/.pdf files into per-paper subdirectories. Added Antje Worring as co-author on all papers. Synced PDFs to pdfs/ for download site.
54 lines
2.5 KiB
TeX
54 lines
2.5 KiB
TeX
\section{Security Considerations}
|
|
|
|
\label{sec:security}
|
|
|
|
\subsection{Long-Range Attacks}
|
|
|
|
Lux is immune to long-range attacks because finality is based on the live stake distribution
|
|
at the time of validation. An attacker cannot rewrite history without controlling the staking
|
|
keys at the original validation time. Unlike Nakamoto consensus, old private keys cannot
|
|
generate alternative histories.
|
|
|
|
\subsection{Eclipse Attacks}
|
|
|
|
Lux mitigates eclipse attacks through:
|
|
\begin{enumerate}[nosep]
|
|
\item Randomized peer selection with VRF-based self-certification
|
|
\item Diverse outbound connections across geographic regions
|
|
\item IP diversity requirements: the P-Chain limits validators per /24 prefix
|
|
\item Stake-weighted sampling ensures economic cost for eclipsing high-stake validators
|
|
\end{enumerate}
|
|
|
|
\subsection{Sybil Attacks}
|
|
|
|
Sybil resistance is provided by Proof-of-Stake: creating validators requires staking LUX.
|
|
The minimum stake of 2,000 LUX sets a floor proportional to market price. For a 33\%
|
|
attack at current valuations, the cost exceeds \$100 million USD.
|
|
|
|
\subsection{Timing Attacks}
|
|
|
|
An adversary may delay messages to disrupt the consecutive counter. After GST, delays are
|
|
bounded by $\Delta$. Before GST, the adversary can stall progress but cannot cause safety
|
|
violations, since all honest nodes remain in the same metastable state.
|
|
|
|
\subsection{Grinding Attacks on VRF Tie-Breaking}
|
|
|
|
An adversary controlling multiple VRF keys could attempt to ``grind'' favorable tie-breaking
|
|
outcomes by choosing which keys to use. Since the VRF input includes the round number
|
|
(committed before key selection), and VRF outputs are deterministic per key, the adversary's
|
|
advantage is limited to choosing among $f$ possible outputs, providing at most $\log_2 f$
|
|
bits of advantage. For $f = n/3 \approx 333$, this is $\approx 8.4$ bits, negligible compared
|
|
to the $\beta k D(\alpha \| 1-\alpha) \approx 333$ bits of security margin.
|
|
|
|
\subsection{Adaptive Adversaries}
|
|
|
|
Our security model assumes a static adversary. Against an adaptive adversary that can corrupt
|
|
validators during execution, the protocol's security depends on the corruption delay $\delta_c$.
|
|
If $\delta_c > \beta \cdot \Delta_{\text{round}}$, the adversary cannot corrupt enough
|
|
validators within a finalization window to violate safety. We leave formal analysis of
|
|
adaptive adversaries to future work.
|
|
|
|
%% -----------------------------------------------------------------------
|
|
%% 10. FORMAL VERIFICATION
|
|
%% -----------------------------------------------------------------------
|