PITTSBURGH — In a major development for the fields of cryptography and data privacy, researchers Andrew Krapivin and Elaine Shi from Carnegie Mellon University have published a new scientific paper that shatters a long-standing efficiency barrier in the design of secure data structures. Their research successfully overcomes what the academic community has termed the "log-squared curse," presenting optimal solutions for both statistically secure Oblivious RAM (ORAM) and oblivious search trees within the external-memory model.
The findings, detailed in a paper titled "Optimal Statistical Oblivious RAM and Search Tree in the External-Memory Model" (Paper 2026/2419) and released on the Cryptology ePrint Archive, address fundamental theoretical limitations that have constrained cryptographic memory access protocols for years. By achieving optimal performance bounds under standard and mild parameter assumptions, the work brings practical, high-security oblivious algorithms closer to mainstream systems architecture.
The Privacy Challenge of Untrusted Memory
In modern computing, data security is no longer confined to isolated local environments. Cloud computing, outsourced data storage, and hardware-enclave architectures require clients to store sensitive information in memory spaces that may be managed by untrusted third parties. Even when data is fully encrypted at rest, the pattern in which a client accesses that data—such as the memory addresses read or written during a search or update operation—can leak profound amounts of confidential information. An observant adversary monitoring these access patterns can potentially deduce database contents, cryptographic keys, or user behavior profiles through side-channel analyses.
To neutralize this vulnerability, researchers developed Oblivious RAM, or ORAM. An ORAM scheme allows a trusted client to interact with encrypted data stored in untrusted memory in such a way that the sequence of memory access patterns reveals absolutely nothing about the underlying data or the nature of the queries being executed.
However, achieving this high degree of privacy has historically carried a steep performance penalty. It is a well-established theoretical principle in computer science that any ORAM scheme must suffer from a logarithmic overhead. Specifically, each memory request must incur a lower bound of $Omega(log_B N)$ block transfers on average, where $N$ represents the size of the logical memory and $B$ denotes the block size of the oblivious implementation. This same fundamental lower bound extends naturally to oblivious search trees, which are essential data structures used to organize, sort, and retrieve information efficiently.
The Long-Standing Curse of Log-Squared Overhead
Despite the theoretical inevitability of the logarithmic lower bound, actual protocol designs have long fallen short of this ideal performance threshold. For years, researchers and engineers have been stuck grappling with a persistent efficiency penalty known as the "log-squared curse," which has manifested conspicuously across two distinct fronts in the literature.
The first front involves statistically secure ORAMs. While researchers have successfully established optimality for computationally secure ORAM schemes under specific conditions—such as when the block size $B = Theta(1)$ or when $B geq log^1+epsilon N$ for an arbitrarily small positive constant $epsilon$—progress regarding statistical security has remained largely stagnant. Computational security relies on the assumption that an adversary has limited computational power, whereas statistical security provides a much stronger guarantee that holds true even against adversaries with unbounded computing resources, bounded only by a negligible failure probability.
The second front concerns oblivious search trees. Across the board, researchers have not known how to overcome the log-squared curse for search trees even when relying on standard computational assumptions. Consequently, executing search tree operations obliviously has historically required significantly more block transfers than non-oblivious counterparts, creating a notable performance bottleneck for systems that require both privacy and rapid data navigation.
Breaking the Barrier Across Broad Parameter Regimes
In their new paper, Krapivin and Shi effectively dismantle this dual-front efficiency curse, demonstrating that optimal performance is achievable across broad and practical parameter regimes. Their work yields two primary technical breakthroughs.
The first major result introduces an optimal, statistically secure ORAM scheme that achieves the theoretical lower bound cost of $O(log_B N)$, provided that the block size satisfies the condition $B geq log^1+epsilon N$. This matches the efficiency standards previously restricted to computational security frameworks while upgrading the security guarantee to a statistical level.
The second major result presents an optimal, statistically secure oblivious search tree that similarly achieves an $O(log_B N)$ cost when $B geq log^2+epsilon N$. Remarkably, this performance matches the efficiency of the best known traditional, non-oblivious search trees within this parameter region. In practical terms, this signifies that for search trees under these conditions, achieving complete algorithmic obliviousness comes essentially free of additional asymptotic overhead.
Alignment with Real-World Hardware and System Architectures
A crucial aspect of the Carnegie Mellon researchers’ work is that the assumptions placed on the block size $B$ to achieve these optimal bounds are neither restrictive nor artificial. Instead, they are standard and mild, fitting neatly into established paradigms of both classical computer science and modern hardware design.
The required parameter assumptions coincide directly with the wide cache-line assumption that has long been prevalent in the classical algorithms and external-memory literature, where systems are designed to transfer data in larger blocks to optimize Input/Output efficiency. Furthermore, these conditions align exceptionally well with contemporary hardware-enclave architectures that deploy oblivious algorithms in practical, real-world privacy-preserving systems. Prominent technology deployments, including secure messaging implementations by platforms such as Signal and Meta, rely heavily on trusted hardware environments where minimizing block transfer overhead is vital for maintaining low latency and high throughput.
By resolving theoretical bottlenecks that have persisted in cryptographic literature, Krapivin and Shi’s paper marks a significant milestone in the ongoing effort to make mathematically rigorous data privacy efficient enough for ubiquitous deployment in large-scale modern systems.
