International Association
for Cryptologic Research

Succinct non-interactive arguments of knowledge (SNARKs) are variants of non-interactive zero-knowledge proofs (NIZKs) in which complex statements can be proven in a compact way. SNARKs have had tremendous impact in several areas of cryptography, including verifiable computing, blockchains, and anonymous communication. A recurring concept in many applications is the concept of recursive SNARKs, in which a proof references a previous proof to show an evolved statement. In this work, we investigate malleable SNARKs, a generalization of this concept of recursion. An adaptation of the existing concept of malleable NIZKs, malleable SNARKs allow to modify SNARK proofs to show related statements, but such that such mauled proofs are indistinguishable from “properly generated” fresh proofs of the related statement. We show how to instantiate malleable SNARKs for universal languages and relations, and give a number of applications: the first post-quantum RCCA-secure rerandomizable and updatable encryption schemes, a generic construction of reverse firewalls, and an unlinkable (i.e., computation-hiding) targeted malleable homomorphic encryption scheme. Technically, our malleable SNARK construction relies on recursive proofs, but with a twist: in order to support the strong indistinguishability properties of mauled and fresh SNARK proofs, we need to allow an unbounded recursion depth. To still allow for a reasonable notion of extractability in this setting (and in particular to guarantee that extraction eventually finishes with a “proper” witness that does not refer to a previous SNARK proof), we rely on a new and generic computational primitive called adversarial one-way function (AOWF) that may be of independent interest. We give an AOWF candidate and prove it secure in the random oracle model.

We study blockchain-based provably anonymous payment systems between \emph{light clients}. Such clients interact with the blockchain through full nodes, which can see what the light clients read and write. The goal of our work is to enable light clients to perform anonymous payments, while maintaining privacy even against the full nodes through which they interact with the blockchain. We formalize the problem in the UC model and present a provably secure solution. We show that a variation of tree ORAM gives obliviousness even when an adversary can follow how its own data elements move in the tree. We use this for anonymity via shuffling of payments on the blockchain, while at the same time allowing the light client to know a few positions among which to find its payment without knowing the current state of the blockchain. In comparison to existing works, we are the first ones that simultaneously provide strong anonymity guarantees, provable security, and anonymity with respect to full nodes. Along the way, we make several contributions that may be of independent interest. We define and construct anonymous-coin friendly encryption schemes and show how they can be used within anonymous payment systems. We define and construct efficient compressible randomness beacons, which produce unpredictable values in regular intervals and allow for storing all published values in a short digest.

Minimizing the round complexity of byzantine broadcast is a fundamental question in distributed computing and cryptography. In this work, we present the first \emph{early stopping} byzantine broadcast protocol that tolerates up to $t=n-1$ malicious corruptions and terminates in $\cO(\min\{f^2,t+1\})$ rounds for any execution with $f\leq t$ \emph{actual corruptions}. Our protocol is deterministic, adaptively secure, and works assuming a plain public key infrastructure. Prior early-stopping protocols all either require honest majority or tolerate only up to $t=(1-\epsilon)n$ malicious corruptions while requiring either trusted setup or strong number theoretic hardness assumptions. As our key contribution, we show a novel tool called a \emph{polarizer} that allows us to transfer certificate-based strategies from the honest majority setting to settings with a dishonest majority.

Recently, time-based primitives such as time-lock puzzles (TLPs) and verifiable delay functions (VDFs) have received a lot of attention due to their power as building blocks for cryptographic protocols. However, even though exciting improvements on their efficiency and security (\textit{e.g.} achieving non-malleability) have been made, most of the existing constructions do not offer general composability guarantees and thus have limited applicability. Baum \textit{et al.} (EUROCRYPT 2021) presented in TARDIS the first (im)possibility results on constructing TLPs with Universally Composable (UC) security and an application to secure two-party computation with output-independent abort (OIA-2PC), where an adversary has to decide to abort \emph{before} learning the output. While these results establish the feasibility of UC-secure TLPs and applications, they are limited to the two-party scenario and suffer from complexity overheads. In this paper, we introduce the first UC constructions of VDFs and of the related notion of publicly verifiable TLPs (PV-TLPs). We use our new UC VDF to prove a folklore result on VDF-based randomness beacons used in industry and build an improved randomness beacon from our new UC PV-TLPs. We moreover construct the first multiparty computation protocol with punishable output-independent aborts (POIA-MPC), \textit{i.e.} MPC with OIA and financial punishment for cheating. Our novel POIA-MPC both establishes the feasibility of (non-punishable) OIA-MPC and significantly improves on the efficiency of state-of-the-art OIA-2PC and (non-OIA) MPC with punishable aborts.

In his landmark paper at TCC 2008 Paul Valiant introduced the notion of ``incrementally verifiable computation'' which enables a prover to incrementally compute a succinct proof of correct execution of a (potentially) long running process. The paper later won the 2019 TCC test of time award. The construction was proven secure in the random oracle model without any further computational assumptions. However, the overall proof was given using a non-standard version of the random-oracle methodology where sometimes the hash function is a random oracle and sometimes it has a short description as a circuit. Valiant clearly noted that this model is non-standard, but conjectured that the standard random oracle methodology would not suffice. This conjecture has been open for 14 years. We prove that if the proof system can receive a long witness as input in an incremental manner and is also zero-knowledge then the conjecture is true. Valiant's original construction does not have these properties but can easily be extended to have them in his model. We relate our result to recent possibility and impossibility results for SNARKs and incrementally verifiable computation.

Subversion attacks undermine security of cryptographic protocols by replacing a legitimate honest party's implementation with one that leaks information in an undetectable manner. An important limitation of all currently known techniques for designing cryptographic protocols with security against subversion attacks is that they do not automatically guarantee security in the realistic setting where a protocol session may run concurrently with other protocols. We remedy this situation by providing a foundation of reverse firewalls (Mironov and Stephens-Davidowitz, EUROCRYPT'15) in the universal composability (UC) framework (Canetti, FOCS'01 and J. ACM'20). More in details, our contributions are threefold: - We generalize the UC framework to the setting where each party consists of a core (which has secret inputs and is in charge of generating protocol messages) and a firewall (which has no secrets and sanitizes the outgoing/incoming communication from/to the core). Both the core and the firewall can be subject to different flavors of corruption, modeling different kinds of subversion attacks. For instance, we capture the setting where a subverted core looks like the honest core to any efficient test, yet it may leak secret information via covert channels (which we call specious subversion). - We show how to sanitize UC commitments and UC coin tossing against specious subversion, under the DDH assumption. - We show how to sanitize the classical GMW compiler (Goldreich, Micali and Wigderson, STOC 1987) for turning MPC with security in the presence of semi-honest adversaries into MPC with security in the presence of malicious adversaries. This yields a completeness theorem for maliciously secure MPC in the presence of specious subversion. Additionally, all our sanitized protocols are transparent, in the sense that communicating with a sanitized core looks indistinguishable from communicating with an honest core. Thanks to the composition theorem, our methodology allows, for the first time, to design subversion-resilient protocols by sanitizing different sub-components in a modular way.

Many decentralized systems rely on flooding protocols for message dissemination. In such a protocol, the sender of a message sends it to a randomly selected set of peers. These peers again send the message to their randomly selected peers, until every network participant has received the message. This type of protocols clearly fail in face of an adaptive adversary who can simply corrupt all peers of the sender and thereby prevent the message from being delivered. Nevertheless, flooding protocols are commonly used within protocols that aim to be cryptographically secure, most notably in blockchain protocols. While it is possible to revert to static corruptions, this gives unsatisfactory security guarantees, especially in the setting of a blockchain that is supposed to run for an extended period of time. To be able to provide meaningful security guarantees in such settings, we give precise semantics to what we call $\delta$-delayed adversaries in the Universal Composability (UC) framework. Such adversaries can adaptively corrupt parties, but there is a delay of time $\delta$ from when an adversary decides to corrupt a party until they succeed in overtaking control of the party. Within this model, we formally prove the intuitive result that flooding protocols are secure against $\delta$-delayed adversaries when $\delta$ is at least the time it takes to send a message from one peer to another plus the time it takes the recipient to resend the message. To this end, we show how to reduce the adaptive setting with a $\delta$-delayed adversary to a static experiment with an Erdős–Rényi graph. Using the established theory of Erdős–Rényi graphs, we provide upper bounds on the propagation time of the flooding functionality for different neighborhood sizes of the gossip network. More concretely, we show the following for security parameter $\kappa$, point-to-point channels with delay at most $\Delta$, and $n$ parties in total, with a sufficiently delayed adversary that can corrupt any constant fraction of the parties: If all parties send to $\Omega(\kappa)$ parties on average, then we can realize a flooding functionality with maximal delay $\mathcal{O}\bigl(\Delta \cdot \log (n) \bigr)$; and if all parties send to $\Omega\bigl( \sqrt{\kappa n \log (n)} \bigr)$ parties on average, we can realize a flooding functionality with maximal delay $\mathcal{O}(\Delta)$.

We distill a simple information-theoretic model for randomness extraction motivated by the task of generating publicly verifiable randomness in blockchain settings and which is closely related to You-Only-Speak-Once (YOSO) protocols (CRYPTO 2021). With the goal of avoiding denial-of-service attacks, parties speak only once and in sequence by broadcasting a public value and forwarding secret values to future parties. Additionally, an unbounded adversary can corrupt any chosen subset of at most t parties. In contrast, existing YOSO protocols only handle random corruptions. As a notable example, considering worst-case corruptions allows us to reduce trust in the role assignment mechanism, which is assumed to be perfectly random in YOSO. We study the maximum corruption threshold t which allows for unconditional randomness extraction in our model: - With respect to feasibility, we give protocols for t corruptions and n=6t+1 or n=5t parties depending on whether the adversary learns secret values forwarded to corrupted parties immediately once they are sent or only once the corrupted party is executed, respectively. Both settings are motivated by practical implementations of secret value forwarding. To design such protocols, we go beyond the committee-based approach that is sufficient for random corruptions in YOSO but turns out to be sub-optimal for chosen corruptions. - To complement our protocols, we show that low-error randomness extraction is impossible with corruption threshold t and n\leq 4t parties in both settings above. This also provides a separation between chosen and random corruptions, since the latter allows for randomness extraction with close to n/2 random corruptions.

A number of recent works have constructed cryptographic protocols with flavors of adaptive security by having a randomly-chosen anonymous committee run at each round. Since most of these protocols are stateful, transferring secret states from past committees to future, but still unknown, committees is a crucial challenge. Previous works have tackled this problem with approaches tailor-made for their specific setting, which mostly rely on using a blockchain to orchestrate auxiliary committees that aid in the state hand-over process. In this work, we look at this challenge as an important problem on its own and initiate the study of Encryption to the Future (EtF) as a cryptographic primitive. First, we define a notion of an EtF scheme where time is determined with respect to an underlying blockchain and a lottery selects parties to receive a secret message at some point in the future. While this notion seems overly restrictive, we establish two important facts: 1. if used to encrypt towards parties selected in the “far future”, EtF implies witness encryption for NP over a blockchain; 2. if used to encrypt only towards parties selected in the “near future”, EtF is not only sufficient for transferring state among committees as required by previous works, but also captures previous tailor-made solutions. To corroborate these results, we provide a novel construction of EtF based on witness encryption over commitments (cWE), which we instantiate from a number of standard assumptions via a construction based on generic cryptographic primitives. Finally, we show how to use “near future” EtF to obtain “far future” EtF with a protocol based on an auxiliary committee whose communication complexity is independent of the length of plaintext messages being sent to the future.

We present a unified view of the two-party and multi-party computation protocols based on oblivious transfer first outlined in Nielsen et al. (CRYPTO 2012) and Larraia et al. (CRYPTO 2014). We present a number of modifications and improvements to these earlier presentations, as well as full proofs of the entire protocol. Improvements include a unified pre-processing and online MAC methodology, mechanisms to pass between different MAC variants and fixing a minor bug in the protocol of Larraia et al. in relation to a selective failure attack. It also fixes a minor bug in Nielsen et al. resulting from using Jensen’s inequality in the wrong direction in an analysis.

Time-based primitives like time-lock puzzles (TLP) are finding widespread use in practical protocols, partially due to the surge of interest in the blockchain space where TLPs and related primitives are perceived to solve many problems. Unfortunately, the security claims are often shaky or plainly wrong since these primitives are used under composition. One reason is that TLPs are inherently not UC secure and time is tricky to model and use in the UC model. On the other hand, just specifying standalone notions of the intended task, left alone correctly using standalone notions like non-malleable TLPs only, might be hard or impossible for the given task. And even when possible a standalone secure primitive is harder to apply securely in practice afterwards as its behavior under composition is unclear. The ideal solution would be a model of TLPs in the UC framework to allow simple modular proofs. In this paper we provide a foundation for proving composable security of practical protocols using time-lock puzzles and related timed primitives in the UC model. We construct UC-secure TLPs based on random oracles and show that using random oracles is necessary. In order to prove security, we provide a simple and abstract way to reason about time in UC protocols. Finally, we demonstrate the usefulness of this foundation by constructing applications that are interesting in their own right, such as UC-secure two-party computation with output-independent abort.

The inherent difficulty of maintaining stateful environments over long periods of time gave rise to the paradigm of serverless computing, where mostly-stateless components are deployed on demand to handle computation tasks, and are teared down once their task is complete. Serverless architecture could offer the added benefit of improved resistance to targeted denial-of-service attacks, by hiding from the attacker the physical machines involved in the protocol until after they complete their work. Realizing such protection, however, requires that the protocol only uses stateless parties, where each party sends only one message and never needs to speaks again. Perhaps the most famous example of this style of protocols is the Nakamoto consensus protocol used in Bitcoin: A peer can win the right to produce the next block by running a local lottery (mining), all while staying covert. Once the right has been won, it is executed by sending a single message. After that, the physical entity never needs to send more messages. We refer to this as the You-Only-Speak-Once (YOSO) property, and initiate the formal study of it within a new model that we call the YOSO model. Our model is centered around the notion of roles, which are stateless parties that can only send a single message. Crucially, our modelling separates the protocol design, that only uses roles, from the role-assignment mechanism, that assigns roles to actual physical entities. This separation enables studying these two aspects separately, and our YOSO model in this work only deals with the protocol-design aspect. We describe several techniques for achieving YOSO MPC; both computational and information theoretic. Our protocols are synchronous and provide guaranteed output delivery (which is important for application domains such as blockchains), assuming honest majority of roles in every time step. We describe a practically efficient computationally-secure protocol, as well as a proof-of-concept information theoretically secure protocol.

Private information retrieval (PIR) lets a client retrieve an entry from a database without the server learning which entry was retrieved. Here we study a weaker variant that we call random-index PIR (RPIR), where the retrieved index is an output rather than an input of the protocol, and is chosen at random. RPIR is clearly weaker than PIR, but it suffices for some interesting applications and may be realized more efficiently than full-blown PIR. We report here on two lines of work, both tied to RPIR but otherwise largely unrelated. The first line of work studies RPIR as a primitive on its own. Perhaps surprisingly, we show that RPIR is in fact equivalent to PIR when there are no restrictions on the number of communication rounds. On the other hand, RPIR can be implemented in a “noninteractive” setting (with preprocessing), which is clearly impossible for PIR. For two-server RPIR we show a truly noninteractive solution, offering information-theoretic security without any pre-processing. The other line of work, which was the original motivation for our work, uses RPIR to improve on the recent work of Benhamouda et al. (TCC’20) for maintaining secret values on public blockchains. Their solution depends on a method for selecting many random public keys from a PKI while hiding most of the selected keys from an adversary. However, the method they proposed is vulnerable to a double-dipping attack, limiting its resilience. Here we observe that an RPIR protocol, where the client is implemented via secure MPC, can eliminate that vulnerability. We thus get a secrets-on-blockchain protocol (and more generally large-scale MPC), resilient to any fraction f < 1/2 of corrupted parties, resolving the main open problem left from the work of Benhamouda et al. As the client in this solution is implemented via secure MPC, it really brings home the need to make it as efficient as possible. We thus strive to explore whatever efficiency gains we can get by using RPIR rather than PIR. We achieve more gains by using batch RPIR where multiple indexes are retrieved at once. Lastly, we observe that this application can make do with a weaker security guarantee than full RPIR, and show that this weaker variant can be realized even more efficiently. We discuss one protocol in particular, that may be attractive for practical implementations.

Threshold secret sharing allows a dealer to split a secret into $n$ shares such that any authorized subset of cardinality at least $t$ of those shares efficiently reveals the secret, while at the same time any unauthorized subset of cardinality less than $t$ contains no information about the secret. Leakage-resilience additionally requires that the secret remains hidden even if one is given a bounded amount of additional leakage from every share. In this work, we study leakage-resilient secret sharing schemes and prove a lower bound on the share size and the required amount randomness of any information-theoretically secure scheme. We prove that for any information-theoretically secure leakage-resilient secret sharing scheme either the amount of randomness across all shares or the share size has to be linear in $n$. More concretely, for a secret sharing scheme with $p$-bit long shares, $\ell$-bit leakage per share, where $\widehat{t}$ shares uniquely define the remaining $n - \widehat{t}$ shares, it has to hold that $p \ge \frac{\ell (n - t)}{\widehat{t}}$. We use this lower bound to gain further insights into a question that was recently posed by Benhamouda et al. (CRYPTO'18), who ask to what extend existing regular secret sharing schemes already provide protection against leakage. The authors proved that Shamir's secret sharing is $1$-bit leakage-resilient for reconstruction thresholds $t \geq 0.85n$ and conjectured that it is also $1$-bit leakage-resilient for any other threshold that is a constant fraction of the total number of shares. We do not disprove their conjecture, but show that it is the best one could possibly hope for. Concretely, we show that for large enough $n$ and any constant $0< c < 1$ it holds that Shamir's secret sharing scheme is \emph{not} leakage-resilient for $t \leq \frac{cn}{\log n}$. In contrast to the setting with information-theoretic security, we show that our lower bound does not hold in the computational setting. That is, we show how to construct a leakage-resilient secret sharing scheme in the random oracle model that is secure against computationally bounded adversaries and violates the lower bound stated above.

Reverse firewalls were introduced at Eurocrypt 2015 by Miro-nov and Stephens-Davidowitz, as a method for protecting cryptographic protocols against attacks on the devices of the honest parties. In a nutshell: a reverse firewall is placed outside of a device and its goal is to ``sanitize'' the messages sent by it, in such a way that a malicious device cannot leak its secrets to the outside world. It is typically assumed that the cryptographic devices are attacked in a ``functionality-preserving way'' (i.e.~informally speaking, the functionality of the protocol remains unchanged under this attacks). In their paper, Mironov and Stephens-Davidowitz construct a protocol for passively-secure two-party computations with firewalls, leaving extension of this result to stronger models as an open question. In this paper, we address this problem by constructing a protocol for secure computation with firewalls that has two main advantages over the original protocol from Eurocrypt 2015. Firstly, it is a \emph{multi}party computation protocol (i.e.~it works for an arbitrary number $n$ of the parties, and not just for $2$). Secondly, it is secure in much stronger corruption settings, namely in the \emph{actively corruption model}. More precisely: we consider an adversary that can fully corrupt up to $n-1$ parties, while the remaining parties are corrupt in a functionality-preserving way. Our core techniques are: malleable commitments and malleable non-interactive zero-knowledge, which in particular allow us to create a novel protocol for multiparty augmented coin-tossing into the well with reverse firewalls (that is based on a protocol of Lindell from Crypto 2001).

Non-malleable codes (Dziembowski et al., ICS’10 and J. ACM’18) are a natural relaxation of error correcting/detecting codes with useful applications in cryptography. Informally, a code is non-malleable if an adversary trying to tamper with an encoding of a message can only leave it unchanged or modify it to the encoding of an unrelated value. This paper introduces continuous non-malleability, a generalization of standard non-malleability where the adversary is allowed to tamper continuously with the same encoding. This is in contrast to the standard definition of non-malleable codes, where the adversary can only tamper a single time. The only restriction is that after the first invalid codeword is ever generated, a special self-destruct mechanism is triggered and no further tampering is allowed; this restriction can easily be shown to be necessary. We focus on the split-state model, where an encoding consists of two parts and the tampering functions can be arbitrary as long as they act independently on each part. Our main contributions are outlined below. We show that continuous non-malleability in the split-state model is impossible without relying on computational assumptions. We construct a computationally secure split-state code satisfying continuous non-malleability in the common reference string (CRS) model. Our scheme can be instantiated assuming the existence of collision-resistant hash functions and (doubly enhanced) trapdoor permutations, but we also give concrete instantiations based on standard number-theoretic assumptions. We revisit the application of non-malleable codes to protecting arbitrary cryptographic primitives against related-key attacks. Previous applications of non-malleable codes in this setting required perfect erasures and the adversary to be restricted in memory. We show that continuously non-malleable codes allow to avoid these restrictions.

Non-malleable codes (NMCs), introduced by Dziembowski, Pietrzak and Wichs [20], provide a useful message integrity guarantee in situations where traditional error-correction (and even error-detection) is impossible; for example, when the attacker can completely overwrite the encoded message. NMCs have emerged as a fundamental object at the intersection of coding theory and cryptography. In particular, progress in the study of non-malleable codes and the related notion of non-malleable extractors has led to new insights and progress on even more fundamental problems like the construction of multi-source randomness extractors. A large body of the recent work has focused on various constructions of non-malleable codes in the split-state model. Many variants of NMCs have been introduced in the literature, e.g., strong NMCs, super strong NMCs and continuous NMCs. The most general, and hence also the most useful notion among these is that of continuous non-malleable codes, that allows for continuous tampering by the adversary. We present the first efficient information-theoretically secure continuously non-malleable code in the constant split-state model. We believe that our main technical result could be of independent interest and some of the ideas could in future be used to make progress on other related questions.

We prove a lower bound on the communication complexity of unconditionally secure multiparty computation, both in the standard model with $$n=2t+1$$ parties of which t are corrupted, and in the preprocessing model with $$n=t+1$$ . In both cases, we show that for any $$g \in \mathbb {N}$$ there exists a Boolean circuit C with g gates, where any secure protocol implementing C must communicate $$\varOmega (n g)$$ bits, even if only passive and statistical security is required. The results easily extends to constructing similar circuits over any fixed finite field. This shows that for all sizes of circuits, the O(n) overhead of all known protocols when t is maximal is inherent. It also shows that security comes at a price: the circuit we consider could namely be computed among n parties with communication only O(g) bits if no security was required. Our results extend to the case where the threshold t is suboptimal. For the honest majority case, this shows that the known optimizations via packed secret-sharing can only be obtained if one accepts that the threshold is $$t= (1/2 - c)n$$ for a constant c. For the honest majority case, we also show an upper bound that matches the lower bound up to a constant factor (existing upper bounds are a factor $$\lg n$$ off for Boolean circuits).

In this work we present a collection of compilers that take secret sharing schemes for an arbitrary access structure as input and produce either leakage-resilient or non-malleable secret sharing schemes for the same access structure. A leakage-resilient secret sharing scheme hides the secret from an adversary, who has access to an unqualified set of shares, even if the adversary additionally obtains some size-bounded leakage from all other secret shares. A non-malleable secret sharing scheme guarantees that a secret that is reconstructed from a set of tampered shares is either equal to the original secret or completely unrelated. To the best of our knowledge we present the first generic compiler for leakage-resilient secret sharing for general access structures. In the case of non-malleable secret sharing, we strengthen previous definitions, provide separations between them, and construct a non-malleable secret sharing scheme for general access structures that fulfills the strongest definition with respect to independent share tampering functions. More precisely, our scheme is secure against concurrent tampering: The adversary is allowed to (non-adaptively) tamper the shares multiple times, and in each tampering attempt can freely choose the qualified set of shares to be used by the reconstruction algorithm to reconstruct the tampered secret. This is a strong analogue of the multiple-tampering setting for split-state non-malleable codes and extractors.We show how to use leakage-resilient and non-malleable secret sharing schemes to construct leakage-resilient and non-malleable threshold signatures. Classical threshold signatures allow to distribute the secret key of a signature scheme among a set of parties, such that certain qualified subsets can sign messages. We construct threshold signature schemes that remain secure even if an adversary leaks from or tampers with all secret shares.

An Oblivious RAM (ORAM) introduced by Goldreich and Ostrovsky [JACM’96] is a (possibly randomized) RAM, for which the memory access pattern reveals no information about the operations performed. The main performance metric of an ORAM is the bandwidth overhead, i.e., the multiplicative factor extra memory blocks that must be accessed to hide the operation sequence. In their seminal paper introducing the ORAM, Goldreich and Ostrovsky proved an amortized $$\varOmega (\lg n)$$ bandwidth overhead lower bound for ORAMs with memory size n. Their lower bound is very strong in the sense that it applies to the “offline” setting in which the ORAM knows the entire sequence of operations ahead of time.However, as pointed out by Boyle and Naor [ITCS’16] in the paper “Is there an oblivious RAM lower bound?”, there are two caveats with the lower bound of Goldreich and Ostrovsky: (1) it only applies to “balls in bins” algorithms, i.e., algorithms where the ORAM may only shuffle blocks around and not apply any sophisticated encoding of the data, and (2), it only applies to statistically secure constructions. Boyle and Naor showed that removing the “balls in bins” assumption would result in super linear lower bounds for sorting circuits, a long standing open problem in circuit complexity. As a way to circumventing this barrier, they also proposed a notion of an “online” ORAM, which is an ORAM that remains secure even if the operations arrive in an online manner. They argued that most known ORAM constructions work in the online setting as well.Our contribution is an $$\varOmega (\lg n)$$ lower bound on the bandwidth overhead of any online ORAM, even if we require only computational security and allow arbitrary representations of data, thus greatly strengthening the lower bound of Goldreich and Ostrovsky in the online setting. Our lower bound applies to ORAMs with memory size n and any word size $$r \ge 1$$ . The bound therefore asymptotically matches the known upper bounds when $$r = \varOmega (\lg ^2 n)$$ .