DNA-Based Key Distribution

• DNA-based key distribution is a method that uses synthetic DNA as a physical entropy source to generate shared one-time pad keys.
• It involves fabricating, partitioning, and sequencing random oligonucleotide pools to produce correlated index-payload pairs essential for secure key generation.
• The approach achieves information-theoretic security by applying randomness extraction techniques like block-5 Purine Parity Digitization (5PPD) and error reconciliation.

Searching arXiv for the cited papers to ground the article in the provided source material. arXiv search: locating the two focal papers and nearby DNA cryptography work. DNA-based key distribution denotes cryptographic schemes in which DNA is used not merely as a metaphorical encoding alphabet, but as a source, representation, or physically distributed substrate of shared secret randomness. In the strictest sense, the topic concerns protocols that enable distant parties to derive the same secret keying material from duplicated DNA matter, later digitized into a One-Time Pad (OTP) mask suitable for information-theoretically secure communication (Jaudou et al., 17 Mar 2026). A broader, but less precise, usage also includes hybrid encryption schemes in which a session key is written as a DNA sequence and then used inside a conventional cryptosystem; such schemes may involve DNA in key usage without actually specifying a complete key-distribution protocol (Benatmane et al., 2023). The distinction between these two senses is central to the literature.

1. Scope and definitional boundaries

A useful technical distinction is between key agreement, key distribution / key transport, and encryption schemes that assume a shared key already exists. In the material considered here, the synchronized-DNA-source approach is an actual key-distribution primitive: Alice manufactures a random molecular pool, duplicates it into two pads, stores one locally, sends the other to Bob, and later both independently sequence their pads to recover correlated digital secrets (Jaudou et al., 17 Mar 2026). By contrast, the Rabin–OTP–Feistel hybrid is mainly a hybrid encryption scheme with an assumed external secret-key sharing mechanism for the DNA OTP key; it does not specify an in-paper key distribution protocol (Benatmane et al., 2023).

Aspect	Synchronized DNA sources	Rabin–OTP–Feistel hybrid
Primary cryptographic role of DNA	Shared entropy source	Session-key representation
Secret derivation model	Duplicated DNA pads later sequenced locally	Sender generates DNA key and says it is sent via an external method
Status as key-distribution protocol	Explicit primitive for later OTP masking	Partial only; key sharing is assumed rather than fully specified

This separation is important because DNA cryptography has often been conflated with message encoding in DNA or symbolic DNA operations. In the synchronized-source model, DNA is the key source. In the Rabin-based hybrid, DNA is a symbolic quaternary representation used to generate a binary OTP-style key, while message encryption itself remains numerical and binary.

2. Molecular construction of synchronized DNA entropy sources

The experimentally demonstrated DNA-based key-distribution primitive uses random pools of synthetic DNA to install a synchronized entropy source between distant parties (Jaudou et al., 17 Mar 2026). Alice first orders two partially random oligonucleotide pools from a commercial supplier: an index pool and a payload pool. Each strand contains random domains produced by degenerate synthesis, where each position is chemically synthesized using a mixture of $A$, $C$, $G$, and $T$.

The experimental design uses 14 random blocks of length $n=5$, separated by fixed 6-nt spacers. The full oligos are 197 bases for the index strand and 195 bases for the payload strand, with complementary 3' ends allowing them to anneal and be polymerase-extended into a longer duplex. Randomness is introduced through three mechanisms emphasized by the authors: random nucleotide incorporation during synthesis; random association between one index strand and one payload strand during duplex assembly; and random bottleneck sampling of a subset of the assembled duplexes. The resulting pool is a set of double-stranded DNA keys, each a unique random index-payload pair present in duplex form.

The synchronized-source property arises because duplication happens before reading. Alice assembles a large random molecular population, then physically partitions or amplifies and splits it into two pads. Since both pads derive from the same underlying molecular sample, Alice and Bob later recover overlapping sets of the same index-payload associations by sequencing locally. The paper makes the role division explicit: the index acts as a public locator, and the payload is the secret attached to that locator.

After consensus formation and filtering, blocks 8–14 are fused into the index and blocks 15–28 into the payload. The index therefore has 6 random blocks of 5 nt, or 30 nt, giving an indexing space of approximately $4^{30}\sim 10^{18}$. The payload spans 14 blocks of 5 nt = 70 nt. The paper also reserves blocks 1–7 for channel/error estimation. For the full random-domain combinatorial space in the chosen oligo design, it gives $2^{140}\approx10^{42}$.

This architecture places DNA-based key distribution between classical pre-shared-key systems and Quantum Key Distribution (QKD). Like a courier-delivered pad, it relies on a physical installation phase. Like QKD, it aims at physics-based rather than computationally hardness-based key establishment and provides some capacity to detect interference. Unlike QKD, the key-generation phase after installation requires no optical or quantum channel.

3. Installation, sequencing, and shared-mask derivation

The protocol has two phases: installation and use (Jaudou et al., 17 Mar 2026). During installation, Alice synthesizes the random oligo pools, assembles duplex keys by annealing index and payload strands and extending them, purifies the resulting 365-bp dsDNA keys, and bottlenecks them to a chosen diversity. In the Paris–Tokyo demonstration she sampled approximately $30\times 10^6$ molecules. She then amplified the sample by PCR, split it into two aliquots, kept one in Paris, and shipped the other to Tokyo, where it was stored about a month before use.

During use, both parties locally sequence their respective pads on nanopore sequencers. In the large experiment, the methods specify PromethION Solo 2 runs on PRO-MIN114 flow cells. Sequencing generated 195.55 million reads in Paris and 178.21 million + 143.42 million reads in the two Tokyo runs. Reads are aligned to a reference with random positions represented by $N$, clustered by the random blocks, and converted into high-quality consensus key sequences with associated quality metrics and cluster sizes.

The public discussion stage is deliberately limited. Bob publicly sends his list of indices to Alice. Alice intersects Bob’s index set with her own, chooses a random ordering $O_r$ for the common indices, and sends this ordered intersection back to Bob. Both sides then arrange their secret payloads according to $C$0. Those ordered payload sequences are digitized into a common binary string, which becomes the OTP mask. The payloads are not transmitted digitally; only index information and ordering are exposed.

The set-theoretic structure is explicit in the attack analysis. Alice and Bob each hold sets of consensus index-payload pairs, and the protocol seeks a large secret overlap while revealing only public locators. In the simple theft-without-replacement scenario, the paper writes

$C$1

This suggests a conceptual shift from “sending a key” toward “installing a shared entropy reservoir” that can be digitized later and in different amounts. The protocol is therefore closer to scalable physical pre-distribution than to online computational key exchange.

4. Randomness extraction, bias suppression, and reconciliation

A central technical issue is that raw synthetic DNA is not unbiased. The paper reports position-dependent base-frequency drift along the synthesis direction and correlations between neighboring bases, with pairwise correlations extending up to length 5 (Jaudou et al., 17 Mar 2026). Because of these biases, naive quaternary encodings such as $C$2, $C$3, $C$4, $C$5 are unsuitable for cryptographic randomness extraction from this source. The authors also reject direct application of the von Neumann debiasing method because it assumes independence.

The proposed extractor is block-5 Purine Parity Digitization (5PPD). For each 5-base random block, the number of purines ($C$6 or $C$7) is counted modulo 2, and that parity bit is taken as the extracted binary symbol. Payloads are divided into 5-base blocks, one bit is produced per block, and the bits are concatenated column-wise across ordered payloads to form the final long mask. Since each payload contains 14 random blocks, each shared payload contributes 14 extracted bits.

After all filtering, Alice retained 26,586,748 high-quality DNA key sequences and Bob retained 27,915,041. Their exact-overlap set contained 22,603,540 indices, about 82% of Alice’s retained set. After 5PPD digitization, this yielded a 316 Mb shared binary mask. The abstract describes the scale as approximately $C$8 Mb, while the main text gives the explicit post-5PPD shared mask size as 316 Mb.

The paper evaluates the extracted binary stream under NIST SP 800-90B, using the conservative min-entropy output defined as the minimum across ten estimators. For the full mask, it reports min-entropy values of 0.9588 on Alice’s side and 0.9604 on Bob’s side, with the minimum dictated by the compression estimator. The authors state that these values are comparable to approved and commercial RNGs, including Hash-DRBG-SHA256 under FIPS 140-3 comparison.

Residual disagreement remains nonzero and must be corrected before reliable OTP use. To estimate channel error, Alice and Bob publicly compare a non-secret 5PPD test region from the index strand. In the experiment, the two test strings differed at 4189 positions out of 157,401,800, corresponding to

$C$9

The actual whole-mask error was measured at approximately

$G$0

The selected reconciliation mechanism is a Bose–Chaudhuri–Hocquenghem (BCH) code. Alice uses the estimated BER to choose BCH parameters so that Bob’s probability of decoding failure is below

$G$1

The demonstration applies the resulting DNA-derived mask to OTP-encrypt and decrypt a $G$2 color image of the Horsehead Nebula of size 130 Mb. In the discussion, the authors estimate an effective throughput of

$G$3

and argue that a single gram of DNA pads could support petabytes of unconditionally secure transmissions over long periods.

5. Security claims, attack models, and operational assumptions

The strongest security claim in the synchronized-source framework is the standard OTP claim: if the final mask is secret, random, as long as the message, and used only once, then ciphertext secrecy is information-theoretic (Jaudou et al., 17 Mar 2026). The paper is explicit that this “unconditional security” refers to the OTP layer, not to every operational step in the biomolecular supply chain. Practical assumptions remain: the DNA pad installation phase must preserve secrecy sufficiently well; classical synchronization messages must be authenticated; local sequencing and digital processing endpoints must not leak the payload; and residual discrepancies must be corrected.

Two explicit adversarial scenarios are analyzed. In scenario 1, Eve steals a fraction of Bob’s pad without replacement and sequences the stolen sample. In the secure variant of the protocol, Alice does not pre-amplify before partitioning; instead she splits after denaturation so each key exists initially in exactly two single strands, one direct and one reverse-complemented. Under this single-copy sharing model, a theft-only attack cannot create tripartite sharing, captured by

$G$4

In scenario 2, Eve steals Bob’s pad, PCR-amplifies it, splits the amplified material, returns one share to Bob, and keeps one share for herself. To detect such interference, the protocol uses Unique Molecular Identifiers (UMIs) before amplification and sequencing. Since UMIs are attached before downstream amplification, deep sequencing can recover the number of distinct original molecules that fed each consensus-key cluster. The paper defines an interference index

$G$5

where $G$6 denotes UMI multiplicity $G$7 and $G$8 the number of clusters with that multiplicity. Statistical difference between native and compromised distributions is assessed with a $G$9 test, and the reported “safety probability” is

$T$0

where $T$1 is the type-I critical level of that test.

Even the most conservative one-copy attack altered the UMI-based statistics detectably. In that case Eve obtained only about 15% of the shared secret, which the authors state could be neutralized with standard privacy amplification. This provides a QKD-like property, not through no-cloning, but through tamper evidence tied to discrete molecular copy numbers.

The paper also describes a lighter authentication mechanism in ordinary use: because the index space scales exponentially with length, Alice can verify that Bob likely possesses the pad if he can produce enough matching indices. It does not present this as a formal information-theoretic authentication proof. A plausible implication is that the system inherits a familiar asymmetry: secrecy of the OTP mask can be information-theoretic, while authenticity of the public discussion remains a classical systems problem.

6. DNA-derived session keys in hybrid cryptosystems

A distinct line of work uses DNA in a much narrower way: as a representation for a session key inside a hybrid encryption pipeline rather than as a full key-distribution substrate. The paper "A New Hybrid Cryptosystem Involving DNA,Rabin, One Time Pad and Fiestel" combines a DNA-derived one-time-pad-like symmetric key, the Rabin public-key cryptosystem, and a simple Feistel-inspired bit reordering (Benatmane et al., 2023).

The sender-side workflow is specified as follows: generate a random DNA key; convert that DNA key into a binary OTP key; obtain the receiver’s Rabin public key $T$2; modify the plaintext by inserting a secret “spy” symbol before each plaintext character; convert characters to ASCII and concatenate values two by two; encrypt each concatenated number using Rabin encryption; convert each Rabin ciphertext block to a 26-bit binary string; XOR each 26-bit block with the binary OTP key; apply a Feistel-inspired reordering on each 26-bit block; and send the resulting ciphertext. On the receiver side, the operations are inverted: undo the Feistel-inspired transformation, XOR with the same binary OTP key, convert 26-bit blocks back to decimal, use the Rabin private key $T$3 to decrypt each block, choose the Rabin root that begins with the agreed “spy” ASCII prefix, remove the spy prefix, and recover plaintext.

Two distinct keys appear. The first is a DNA OTP key, used for only one secure communication session. The sender generates a random DNA sequence, converts it to binary using

$T$4

and uses the resulting random binary key in the XOR step,

$T$5

The second is the Rabin key pair: the receiver generates distinct primes $T$6 and $T$7 such that

$T$8

computes

$T$9

publishes $n=5$0, and keeps $n=5$1 private. Rabin encryption is

$n=5$2

The paper’s relevance to DNA-based key distribution is therefore partial. It explicitly identifies a session key and a public/private Rabin key pair, but it does not actually present a complete key distribution protocol inside the paper. The weak point is key sharing: the sender generates the DNA key, converts it to binary, and then says it is sent to the receiver using the method described in reference [16], without reproducing that method. The paper does not say that the DNA OTP key is encrypted under Rabin. Rabin encrypts the message blocks, not the session key.

This distinction matters because the work is best characterized as an encryption scheme assuming key availability. It is not a true key agreement protocol like Diffie–Hellman, and it is not a fully specified key transport protocol within the paper. It is also overwhelmingly computational/simulated DNA cryptography rather than wet-lab DNA cryptography: the evidence given includes a MATLAB prototype, DNA sequences taken from websites, and symbolic use of DNA strings without synthesis, PCR, hybridization, gel electrophoresis, sequencing, or ligation.

The example based on the plaintext “Success” illustrates the point. With plaintext length $n=5$3, the scheme uses a 91-nucleotide DNA key, which becomes a 182-bit binary key shown as 7 rows of 26 bits. The receiver generates

$n=5$4

and sends only $n=5$5 to the sender. The “spy” is chosen as “$n=5$6” with ASCII value $n=5$7, leading to the modified text

$n=5$8

and pairwise concatenations

$n=5$9

Rabin encryption then yields

$4^{30}\sim 10^{18}$0

The example demonstrates message processing and recovery, but it does not show how the OTP key is securely transmitted, how [16] works, how parties authenticate each other, or how the “spy” is initially shared.

For research on DNA-based key distribution, the significance of this hybrid scheme is therefore mainly negative and clarifying. It shows how a DNA sequence can be mapped into a binary OTP-style key and used in a layered encryption pipeline, but it does not provide a stand-alone DNA key-distribution mechanism. A plausible implication is that a substantial part of the literature labeled “DNA cryptography” concerns DNA-related key usage or symbolic encoding rather than the cryptographic problem of establishing a shared secret itself.