International Association
for Cryptologic Research

There has been significant progress over the past seven years in model reverse engineering (RE) for neural network (NN) hardware. Although there has been systematization of knowledge (SoK) in an overall sense, however, the treatment from the hardware perspective has been far from adequate. To bridge this gap, this paper systematically categorizes the types of NN hardware used prevalently by the industry/academia, and also the model RE attacks/defenses published in each category. Further, we sub-categorize existing NN model RE attacks based on different criteria including the degree of hardware parallelism, threat vectors like side channels, fault-injection, scan-chain attacks, system-level attacks, type of asset under attack, the type of NN, exact versus approximate recovery, etc.

We make important technical observations and identify key open research directions. Subsequently, we discuss the state-of-the-art defenses against NN model RE, identify certain categorization criteria, and compare the existing works based on these criteria. We note significant qualitative gaps for defenses, and suggest recommendations for important open research directions for protection of NN models. Finally, we discuss limitations of existing work in terms of the types of models where security evaluation or defenses were proposed, and suggest open problems in terms of protecting practically expensive model IPs.

In the useful and well studied model of secret-sharing schemes, there are $n$ parties and a dealer, which holds a secret. The dealer applies some randomized algorithm to the secret, resulting in $n$ strings, called shares; it gives the $i$'th share to the $i$'th party. There are two requirements. (1) correctness: some predefined subsets of the parties can jointly reconstruct the secret from their shares, and (2) security: any other set gets no information on the secret. The collection of predefined qualified sets is called an access structure (AS).

This model assumes that the number of parties is known when preparing the shares and giving the shares to the parties; furthermore, the sharing algorithm and the share size are determined by the number of parties, e.g. in the best-known secret-sharing scheme for an arbitrary $n$-party access structure the share size is $1.5^{n}$ by Appelbaum and Nir.

The assumption that the number of parties is known in advance is problematic in many scenarios. Of course, one can take some upper bound on the number of parties. On one hand, if this bound is big, then the share size will be large even if only few parties actually participate in the scheme. On the other hand, if this bound is small, then there is a risk that too many parties will arrive and no further shares can be produced; this will require an expensive re-sharing of the secret and updating all shares (which can be impossible if some parties are temporally off-line). Thus, we need to consider models with an unbounded number of parties.

To address these concrens, Komargodski, Naor, and Yogev defined \emph{evolving secret-sharing schemes} with an unbounded number of parties. In a nutshell, evolving AS's are defined as a monotone collection of finite qualified sets, such that at any time $t$ a set $A\subseteq [t]$ is either qualified or not, depending only on $A$ itself, and not on $t$ (a `global' monotonicity notion).

Quantum secret sharing (QSS) in the standard $n$-party setting, where the secret is an arbitrary quantum state (say, qbit), rather than classical data. In face of recent advancements in quantum computing, this is a natural notion to consider, and has been studied before.

In this work, we explore the natural notion of quantum evolving secret sharing (QESS). While this notion has been studied by Samadder 20', we make several new contributions. (1) The notion of QESS was only implicit in the above work. We formalize this notion (as well as AS's for which it is applicable), and in particular argue that the variant implied by the above work did not require `global monotonicity' of the AS, which was the standard in the evolving secret sharing literature, and appears to be useful for QESS as well. (2) Discuss the applicability and limitations of the notion in the quantum setting that follow from the no-cloning theorem, and make its usability more limited. Yet, we argue that fundamental advantages of the evovling setting, such as keeping parties' shares independent of the total number of parties that arrive can be mantainted in the quantum setting. (3) We characterize the AS's ammenable to construction of QSSS - so called `no cloning' evolving AS's, and point out that this class is not severly restricted relatively to the class of all evolving AS's. On the positive side, our construction combines the compiler of [Smith 00'] with ideas of hybrid secret sharing of [Goyal et. al 23'], to obtain a construction with share size comparable to the best classical linear share complexity of the scheme.

Cryptographic hash functions are said to be the work-horses of modern cryptography. One of the strongest approaches to assess a cryptographic hash function's security is indifferentiability. Informally, indifferentiability measures to what degree the function resembles a random oracle when instantiated with an ideal underlying primitive. However, proving the indifferentiability security of hash functions has been challenging due to complex simulator designs and proof arguments. The Sponge construction is one of the prevalent hashing method used in various systems. The Sponge has been shown to be indifferentiable from a random oracle when initialized with a random permutation.

In this work, we first introduce $\mathsf{GSponge}$, a generalized form of the Sponge construction offering enhanced flexibility in input chaining, field sizes, and padding types. $\mathsf{GSponge}$ not only captures all existing sponge variants but also unveils new, efficient ones. The generic structure of $\mathsf{GSponge}$ facilitates the discovery of two micro-optimizations for already deployed sponges. Firstly, it allows a new padding rule based on zero-padding and domain-separated inputs, saving one full permutation call in certain cases without increasing the generation time of zero-knowledge proofs. Secondly, it allows to absorb up to $\mathsf{c}/2$ more elements (that can save another permutation call for certain message lengths) without compromising the indifferentiability security. These optimizations enhance hashing time for practical use cases such as Merkle-tree hashing and short message processing.

We then propose a new efficient instantiation of $\mathsf{GSponge}$ called $\mathsf{Sponge2}$ capturing these micro-optimizations and provide a formal indifferentiability proof to establish both $\mathsf{Sponge2}$ and $\mathsf{GSponge}$'s security. This proof, simpler than the original for Sponges, offers clarity and ease of understanding for real-world practitioners. Additionally, it is demonstrated that $\mathsf{GSponge}$ can be safely instantiated with permutations defined over large prime fields, a result not previously formally proven.

$\mathrm{SPHINCS^{+}}$ is a post-quantum signature scheme that, at the time of writing, is being standardized as $\mathrm{SLH\text{-}DSA}$. It is the most conservative option for post-quantum signatures, but the original tight proofs of security were flawed—as reported by Kudinov, Kiktenko and Fedorov in 2020. In this work, we formally prove a tight security bound for $\mathrm{SPHINCS^{+}}$ using the EasyCrypt proof assistant, establishing greater confidence in the general security of the scheme and that of the parameter sets considered for standardization. To this end, we reconstruct the tight security proof presented by Hülsing and Kudinov (in 2022) in a modular way. A small but important part of this effort involves a complex argument relating four different games at once, of a form not yet formalized in EasyCrypt (to the best of our knowledge). We describe our approach to overcoming this major challenge, and develop a general formal verification technique aimed at this type of reasoning. Enhancing the set of reusable EasyCrypt artifacts previously produced in the formal verification of stateful hash-based cryptographic constructions, we (1) improve and extend the existing libraries for hash functions and (2) develop new libraries for fundamental concepts related to hash-based cryptographic constructions, including Merkle trees. These enhancements, along with the formal verification technique we develop, further ease future formal verification endeavors in EasyCrypt, especially those concerning hash-based cryptographic constructions.

One of the main issues to deal with for fully homomorphic encryption is the noise growth when operating on ciphertexts. To some extent, this can be controlled thanks to a so-called gadget decomposition. A gadget decomposition typically relies on radix- or CRT-based representations to split elements as vectors of smaller chunks whose inner products with the corresponding gadget vector rebuilds (an approximation of) the original elements. Radix-based gadget decompositions present the advantage of also supporting the approximate setting: for most homomorphic operations, this has a minor impact on the noise propagation but leads to substantial savings in bandwidth, memory requirements and computational costs. A typical use-case is the blind rotation as used for example in the bootstrapping of the TFHE scheme. On the other hand, CRT-based representations are convenient when machine words are too small for directly accommodating the arithmetic on large operands. This arises in two typical cases: (i) in the hardware case with multipliers of restricted size, e.g., 17 bits; (ii) in the software case for ciphertext moduli above, e.g., 128 bits.

This paper presents new CRT-based gadget decompositions for the approximate setting, which combines the advantages of non-exact decompositions with those of CRT-based decompositions. Significantly, it enables certain hardware or software realizations otherwise hardly supported like the two aforementioned cases. In particular, we show that our new gadget decompositions provide implementations of the (programmable) bootstrapping in TFHE relying solely on native arithmetic and offering extra degrees of parallelism.

In this note, we present additional preliminary analysis dedicated to Ascon-Xof and Ascon-Hash [DEMS19].

The IMDEA Software Institute invites applications for a PhD student in the area of Cryptography. The successful candidate will work under the supervision of Dario Fiore on constructions and applications of cryptographic protocols for secure computation. Topics of particular interest include: zero-knowledge proofs, succinct proof systems and verifiable computation, computation on encrypted data.

Who should apply? The ideal candidates have earned (or are in their last year of) a Master's degree in Computer Science, Mathematics or a related discipline, and have a background in Cryptography. Experience in research or implementation of cryptographic protocols will be considered a plus.

Working at IMDEA Software: Ranked among the Europe's top research institutes in Security and Cryptography, the IMDEA Software Institute offers an inspiring and dynamic collaborative environment with a focus on foundations and applications of cryptography. The Institute is located in the vibrant city of Madrid. The institute provides a competitive salary and funding for research-related travel. The working language at the institute is English.

Dates: The position will span the entire duration of doctoral studies. The starting date is flexible from October 2024. The deadline for applications is July 15th, 2024. Review of applications will begin immediately, and continue until the position is filled.

For further information about the application: https://software.imdea.org/careers/2024-06-phd-picocrypt/

Closing date for applications:

Contact: Dario Fiore (dario.fiore (at) imdea.org)

More information: https://software.imdea.org/careers/2024-06-phd-picocrypt/

The Cryptography Research Group at University College Cork (UCC) is looking for two highly motivated Post-Doctoral or Senior Post-Doctoral Researchers in homomorphic encryption, secure multi-party computation and privacy preservation. The researchers will be employed on the Horizon Europe project “SECURED”, aimed at scaling up the secure processing of health data, and will focus on homomorphic encryption, secure multi-party computation, and (de-)anonymisation and how they can be efficiently used in the e-health context. The Principal Investigator of the project in UCC is Dr. Paolo Palmieri.
Candidates should hold a PhD degree in cryptography or related area, with a good track record of publications. Ideally, they will have experience in one or more of the following areas: homomorphic encryption/secure multiparty computation, lattice-based and post-quantum cryptography, differential privacy, and (de-)anonymisation. Candidates with a background in other areas of cryptography/privacy/security, but with a strong interest in homomorphic encryption, anonymity or differential privacy will also be considered. A strong mathematical background is expected, complemented with programming skills. Experience with relevant libraries such as SEAL, HElib etc. is an asset.
The positions are until December 2025, with a possibility of extension subject to availability of funding. The successful candidates will be appointed at Post-Doctoral or Senior Post-Doctoral level depending on their experience and qualifications. A budget for travel, equipment, publications and other research expenses is available as part of the project.
The Cryptography Research Group is led by Dr Paolo Palmieri and consists of 8 researchers at doctoral and post-doctoral level. The hired researcher will be encouraged to collaborate with other members of the group, and mentor some of the more junior researchers. There will also be ample opportunities to work with other partners in the SECURED project (including some of the top research groups in cryptography, both in industry and academia), as well as with the group’s extensive network of international collaborations.

Closing date for applications:

Contact: Informal inquiries can be made in confidence to Dr Paolo Palmieri, at: p.palmieri@cs.ucc.ie
Applications should be submitted through the University portal at https://ore.ucc.ie/ (search for reference number: 077671).
Deadline: June 17, 2024 at 12:00 (noon) Irish time.

More information: https://security.ucc.ie/vacancies.html

We are seeking an exceptional candidate for a 2-year postdoctoral position in blockchain research, offering a competitive salary of AU$116,679 per annum plus 17% superannuation. Established in 1850, The University of Sydney (USYD) holds the distinction of being the oldest university in Australia and Oceania. It is ranked 18th globally in the QS World University Rankings 2025, and its Computer Science is ranked the top in Australia (QS 2024). Visit the link below for more details and application instructions.

Closing date for applications:

Contact: Jiangshan Yu

More information: https://usyd.wd3.myworkdayjobs.com/en-US/USYD_EXTERNAL_CAREER_SITE/job/Postdoctoral-Research-Fellow---Blockchain_0119398

The research group for Cryptographic Protocols located at the University of Luxembourg and the KASTEL Security Research Labs (Germany) is looking for a PhD student working on cryptographic primitives and protocols enabling privacy, accountability, and transparency.

A background in provable security (e.g., successfully attended courses or a master’s thesis on the subject) is expected.

The candidate will be based at the University of Luxembourg but also profit from regular visits at and joint research projects with the KASTEL Security Research Labs at KIT, Germany.

The candidate’s research will be dealing with privacy-preserving cryptographic building blocks and protocols for important application scenarios and result in both theoretical contributions (protocol designs, security models and proofs, etc.) and their efficient implementation. Privacy-preserving payments and data analytics, misuse-resistant lawful interception, and anonymous communication are research topics of particular interest to us.

If you are interested in joining our group, please send an email including your CV, transcripts, and two references to andy.rupp@uni.lu. As the position should be filled as soon as possible, your application will be considered promptly.

Closing date for applications:

Contact: Andy Rupp (andy.rupp@uni.lu)

This paper presents an optimization of the memory cost of the quantum Information Set Decoding (ISD) algorithm proposed by Bernstein (PQCrypto 2010), obtained by combining Prange's ISD with Grover's quantum search.

When the code has constant rate and length $n$, this algorithm essentially performs a quantum search which, at each iterate, solves a linear system of dimension $\mathcal{O}(n)$. The typical code lengths used in post-quantum public-key cryptosystems range from $10^3$ to $10^5$. Gaussian elimination, which was used in previous works, needs $\mathcal{O}(n^2)$ space to represent the matrix, resulting in millions or billions of (logical) qubits for these schemes.

In this paper, we propose instead to use the algorithm for sparse matrix inversion of Wiedemann (IEEE Trans. inf. theory 1986). The interest of Wiedemann's method is that one relies only on the implementation of a matrix-vector product, where the matrix can be represented in an implicit way. This is the case here.

We propose two main trade-offs, which we have fully implemented, tested on small instances, and benchmarked for larger instances. The first one is a quantum circuit using $\mathcal{O}(n)$ qubits, $\mathcal{O}(n^3)$ Toffoli gates like Gaussian elimination, and depth $\mathcal{O}(n^2 \log n)$. The second one is a quantum circuit using $\mathcal{O}(n \log^2 n)$ qubits, $\mathcal{O}(n^3)$ gates in total but only $\mathcal{O}( n^2 \log^2 n)$ Toffoli gates, which relies on a different representation of the search space.

As an example, for the smallest Classic McEliece parameters we estimate that the Quantum Prange's algorithm can run with 18098 qubits, while previous works would have required at least half a million qubits.

Side channel attacks, and in particular timing attacks, are a fundamental obstacle to obtaining secure implementation of algorithms and cryptographic protocols, and have been widely researched for decades. While cryptographic definitions for the security of cryptographic systems have been well established for decades, none of these accepted definitions take into account the running time information leaked from executing the system. In this work, we give the foundation of new cryptographic definitions for cryptographic systems that take into account information about their leaked running time, focusing mainly on keyed functions such as signature and encryption schemes. Specifically,

(1) We define several cryptographic properties to express the claim that the timing information does not help an adversary to extract sensitive information, e.g. the key or the queries made. We highlight the definition of key-obliviousness, which means that an adversary cannot tell whether it received the timing of the queries with the actual key or the timing of the same queries with a random key.

(2) We present a construction of key-oblivious pseudorandom permutations on a small or medium-sized domain. This construction is not ``fixed-time,'' and at the same time is secure against any number of queries even in case the adversary knows the running time exactly. Our construction, which we call Janus Sometimes Recurse, is a variant of the ``Sometimes Recurse'' shuffle by Morris and Rogaway.

(3) We suggest a new security notion for keyed functions, called noticeable security, and prove that cryptographic schemes that have noticeable security remain secure even when the exact timings are leaked, provided the implementation is key-oblivious. We show that our notion applies to cryptographic signatures, private key encryption and PRPs.

We present an efficient quantum algorithm for solving the semidirect discrete logarithm problem (SDLP) in any finite group. The believed hardness of the semidirect discrete logarithm problem underlies more than a decade of works constructing candidate post-quantum cryptographic algorithms from nonabelian groups. We use a series of reduction results to show that it suffices to consider SDLP in finite simple groups. We then apply the celebrated Classification of Finite Simple Groups to consider each family. The infinite families of finite simple groups admit, in a fairly general setting, linear algebraic attacks providing a reduction to the classical discrete logarithm problem. For the sporadic simple groups, we show that their inherent properties render them unsuitable for cryptographically hard SDLP instances, which we illustrate via a Baby-Step Giant-Step style attack against SDLP in the Monster Group.

Our quantum SDLP algorithm is fully constructive for all but three remaining cases that appear to be gaps in the literature on constructive recognition of groups; for these cases SDLP is no harder than finding a linear representation. We conclude that SDLP is not a suitable post-quantum hardness assumption for any choice of finite group.

A popular way to build post-quantum signature schemes is by first constructing an identification scheme (IDS) and applying the Fiat-Shamir transform to it. In this work we tackle two open questions related to the general applicability of techniques around this approach that together allow for efficient post-quantum signatures with optimal security bounds in the QROM.

First we consider a recent work by Aguilar-Melchor, Hülsing, Joseph, Majenz, Ronen, and Yue (Asiacrypt'23) that showed that an optimal bound for three-round commit & open IDS by Don, Fehr, Majenz, and Schaffner (Crypto'22) can be applied to the five-round Syndrome-Decoding in the Head (SDitH) IDS. For this, they first applied a transform that replaced the first three rounds by one. They left it as an open problem if the same approach applies to other schemes beyond SDitH. We answer this question in the affirmative, generalizing their round-elimination technique and giving a generic security proof for it. Our result applies to any IDS with $2n+1$ rounds for $n>1$. However, a scheme has to be suitable for the resulting bound to not be trivial. We find that IDS are suitable when they have a certain form of special-soundness which many commit & open IDS have.

Second, we consider the hypercube technique by Aguilar-Melchor, Gama, Howe, Hülsing, Joseph, and Yue (Eurocrypt'23). An optimization that was proposed in the context of SDitH and is now used by several of the contenders in the NIST signature on-ramp. It was conjectured that the technique applies generically for the MPC-in-the-Head (MPCitH) technique that is used in the design of many post-quantum IDS if they use an additive secret sharing scheme but this was never proven. In this work we show that the technique generalizes to MPCitH IDS that use an additively homomorphic MPC protocol, and we prove that security is preserved.

We demonstrate the application of our results to the identification scheme of RYDE, a contender in the recent NIST signature on-ramp. While RYDE was already specified with the hypercube technique applied, this gives the first QROM proof for RYDE with an optimally tight bound.

Decentralized payments have evolved from using pseudonymous identifiers to much more elaborate mechanisms to ensure privacy. They can shield the amounts in payments and achieve untraceability, e.g., decoy-based untraceable payments use decoys to obfuscate the actual asset sender or asset receiver. There are two types of decoy-based payments: full decoy set payments that use all other available users as decoys, e.g., Zerocoin, Zerocash, and ZCash, and user-defined decoy set payments where the users select small decoy sets from available users, e.g., Monero, Zether, and QuisQuis.

Existing decoy-based payments face at least two of the following problems: (1) degrading untraceability due to the possibility of payment-graph analysis in user-defined decoy payments, (2) trusted setup, (3) availability issues due to expiring transactions in full decoy sets and epochs, and (4) an ever-growing set of unspent outputs since transactions keep generating outputs without saying which ones are spent. QuisQuis is the first one to solve all these problems; however, QuisQuis requires large cryptographic proofs for validity.

We introduce Nopenena (means ``cannot see''): account-based, confidential, and user-defined decoy set payment protocol, that has short proofs and also avoids these four issues. Additionally, Nopenena can be integrated with zero-knowledge contracts like Zether's $\Sigma-$Bullets and Confidential Integer Processing (CIP) to build decentralized applications. Nopenena payments are about 80% smaller than QuisQuis payments due to Nopenena's novel cryptographic protocol. Therefore, decentralized systems benefit from Nopenena's untraceability and efficiency.

The field of verifiable secret sharing schemes was introduced by Verheul et al. and has evolved over time, including well-known examples by Feldman and Pedersen. Stinson made advancements in combinatorial design-based secret sharing schemes in 2004. Desmedt et al. introduced the concept of frameproofness in 2021, while recent research by Sehrawat et al. in 2021 focuses on LWE-based access structure hiding verifiable secret sharing with malicious-majority settings. Furthermore, Roy et al. combined the concepts of reparable threshold schemes by Stinson et al. and frameproofness by Desmedt et al. in 2023, to develop extendable tensor designs built from balanced incomplete block designs, and also presented a frameproof version of their design. This paper explores ramp-type verifiable secret sharing schemes, and the application of hidden access structures in such cryptographic protocols. Inspired by Sehrawat et al.'s access structure hiding scheme, we develop an $\epsilon$-almost access structure hiding scheme, which is verifiable as well as frameproof. We detail how the concept $\epsilon$-almost hiding is important for incorporating ramp schemes, thus making a fundamental generalisation of this concept.

This note shows practical committing attacks against Rocca-S, an authenticated encryption with associated data scheme designed for 6G applications. Previously, the best complexity of the attack was $2^{64}$ by Derbez et al. in ToSC 2024(1)/FSE 2024. We show that the committing attack against Rocca by Takeuchi et al. in ToSC 2024(2)/FSE 2025 can be applied to Rocca-S, where Rocca is an earlier version of Rocca-S. We show a concrete test vector of our attack. We also point out a committing attack that exploits equivalent keys.

With the increasing demand for Bluetooth devices, various Bluetooth devices support a power-saving mode to reduce power consumption. One of the features of the power-saving mode is that the Bluetooth sessions among devices are temporarily disconnected or close to being disconnected. Prior works have analyzed that the power-saving mode is vulnerable to denial of sleep (DoSL) attacks that interfere with the transition to the power-saving mode of Bluetooth devices, thereby increasing its power consumption. However, to the best of our knowledge, no prior work has analyzed vulnerabilities or attacks on the state after transitioning to the power-saving mode.

To address this issue, we present an attack that abuses two novel vulnerabilities in sleep mode, which is one of the Bluetooth power-saving modes, to break Bluetooth sessions. We name the attack Breaktooth. The attack is the first to abuse the vulnerabilities as an entry point to hijack Bluetooth sessions between victims. The attack also allows overwriting the link key between the victims using the hijacked session, enabling arbitrary command injection on the victims. Furthermore, while many prior attacks assume that attackers can forcibly disconnect the Bluetooth session using methods such as jamming to launch their attacks, our attack does not require such assumptions, making it more realistic.

In this paper, we present the root causes of the Breaktooth attack and their impact. We also provide the technical details of how attackers can secretly detect the sleep mode of their victims. The attackers can easily recognize the state of the victim's Bluetooth session remotely using a standard Linux command. Additionally, we develop a low-cost toolkit to perform our attack and confirm the effectiveness of our attack. Then, we evaluate the attack on 13 types of commodity Bluetooth keyboards and mice that support the sleep mode and show that the attack poses a serious threat to Bluetooth devices supporting the sleep mode. To fix our attack, we present defenses and its proof-of-concept. We responsibly disclosed our findings to the Bluetooth SIG.

The notion of aggregate signatures allows for combining signatures from different parties into a short certificate that attests that *all* parties signed a message. In this work, we lift this notion to capture different, more expressive signing policies. For example, we can certify that a message was signed by a (weighted) threshold of signers.

We present the first constructions of aggregate signatures for monotone policies based on standard polynomial-time cryptographic assumptions. The aggregate signatures in our schemes are succinct, i.e., their size is *independent* of the number of signers. Moreover, verification is also succinct if all parties sign the same message (or if the messages have a succinct representation). All prior work requires either interaction between the parties or non-standard assumptions (that imply SNARKs for NP).

Our signature schemes are based on non-interactive batch arguments (BARGs) for monotone policies [Brakerski-Brodsky-Kalai-Lombardi-Paneth, Crypto'23]. In contrast to previous constructions, our BARGs satisfy a new notion of *adaptive* security which is instrumental to our application. Our new BARGs for monotone policies can be constructed from standard BARGs and other standard assumptions.

Motivated by the problem of detecting AI-generated text, we consider the problem of watermarking the output of language models with provable guarantees. We aim for watermarks which satisfy: (a) undetectability, a cryptographic notion introduced by Christ, Gunn & Zamir (2024) which stipulates that it is computationally hard to distinguish watermarked language model outputs from the model's actual output distribution; and (b) robustness to channels which introduce a constant fraction of adversarial insertions, substitutions, and deletions to the watermarked text. Earlier schemes could only handle stochastic substitutions and deletions, and thus we are aiming for a more natural and appealing robustness guarantee that holds with respect to edit distance.

Our main result is a watermarking scheme which achieves both undetectability and robustness to edits when the alphabet size for the language model is allowed to grow as a polynomial in the security parameter. To derive such a scheme, we follow an approach introduced by Christ & Gunn (2024), which proceeds via first constructing pseudorandom codes satisfying undetectability and robustness properties analogous to those above; our key idea is to handle adversarial insertions and deletions by interpreting the symbols as indices into the codeword, which we call indexing pseudorandom codes. Additionally, our codes rely on weaker computational assumptions than used in previous work. Then we show that there is a generic transformation from such codes over large alphabets to watermarking schemes for arbitrary language models.