Most modern cryptography is built on a gamble. We assume that certain mathematical problems—like factoring large integers or finding discrete logarithms—are too computationally expensive for an adversary to solve in a reasonable timeframe. However, as Jörg Rothe notes in Some Facets of Complexity Theory and Cryptography: A Five-Lectures Tutorial, this computational security is inherently conditional. If an attacker possesses sufficient computing power, or if a mathematical breakthrough (such as quantum computing) invalidates the underlying hardness assumption, the security guarantees vanish entirely. This vulnerability is particularly pressing for emerging infrastructures like 6G networks, where traditional cryptographic primitives face existential threats from quantum acceleration, as discussed by Aitor Brazaola-Vicario et al. in High-Fidelity Coherent-One-Way QKD Simulation Framework for 6G Networks: Bridging Theory and Reality.
Information-theoretic (or unconditional) security offers an alternative paradigm. Instead of making assumptions about an adversary's capabilities, it relies on the mathematical absence of information. If a ciphertext or an observed message contains absolutely no statistical dependency on the underlying secret, an adversary with infinite computational resources and infinite time still cannot extract the secret. The core intuition is simple: you cannot compute what is not there.
Randomization and Secret Sharing
How do we achieve this without the impractical requirement of a one-time pad where the key must be as long as the message? One powerful mechanism is the use of random linear transformations over finite fields. For instance, in Unconditionally Secure Access Control Encryption, Cheuk Ting Li et al. replace traditional public-key primitives with matrix-based randomization to enforce access control policies. Because the transformations map data into a completely uniform distribution from the perspective of unauthorized parties, the architecture remains secure against an all-powerful adversary.
Another foundational approach is secret sharing, where a piece of data is split into multiple shares such that only specific combinations can reconstruct it. Ilan Komargodski et al. explore this in How to Share a Secret, Infinitely, demonstrating how to dynamically scale these access structures while maintaining information-theoretic guarantees.
In database security, Gregory Valiant et al. introduce a method called Secure In Between Accesses (SIBA) in Information Theoretically Secure Databases. By periodically re-randomizing the internal representation of data via XORing secret indices, they ensure that an adversary cannot glean information without exceeding a strict communication threshold, making data theft mathematically impossible regardless of the attacker's processing speed.
Secure Distributed Storage and Integrity Verification
Information-theoretic security is not limited to hiding data; it also guarantees data integrity and possession. In distributed storage systems, data is often broken up and distributed across multiple nodes. Nihar B Shah et al. demonstrate in Information-Theoretically Secure Erasure Codes for Distributed Storage that erasure codes can be designed to protect data from both passive eavesdroppers and active adversaries during node repair operations.
To verify that a remote server is actually storing data without relying on unproven cryptographic assumptions, Mohammad Husain et al. leverage algorithmic information theory and list decoding in Storage Enforcement with Kolmogorov Complexity and List Decoding. By forcing the server to commit physical storage space, the protocol ensures that a proof of possession remains valid against unbounded adversaries.
Furthermore, Giuseppe Ateniese et al. propose in Entangled Cloud Storage a polynomial-based encoding scheme that "entangles" multiple users' files. Because of the global mathematical structure of the polynomial, a cloud provider cannot alter or delete even a tiny fraction of the data without destroying the recovery capability for all users, providing an unconditional integrity guarantee.
Privacy-Preserving Distributed Computation
The rise of collaborative computing and machine learning has shifted the focus of information-theoretic security toward multi-party computation (MPC) and federated learning (FL). In these settings, the goal is to compute a function or train a model over distributed datasets without revealing individual inputs.
For example, Jack K. Fitzsimons et al. apply these principles to contact tracing in A note on blind contact tracing at scale with applications to the COVID-19 pandemic, transforming geographic proximity into an unconditional equality testing problem where exchanged messages contain zero mathematical information about the underlying locations. In the context of machine learning, Basak Guler et al. introduce COPML in A Scalable Approach for Privacy-Preserving Collaborative Machine Learning, combining Lagrange coding with secure multi-party computation to provide a future-proof alternative to traditional encryption.
This paradigm is extensively developed across several facets of federated learning. Maximilian Egger et al. explore secure aggregation in Private Aggregation in Wireless Federated Learning with Heterogeneous Clusters, ensuring that a central server only observes the aggregate sum of updates, completely masking individual gradients against computationally unbounded attackers. The authors extend this framework to hide not only the data but also the server's training objectives in Federated One-Shot Learning with Data Privacy and Objective-Hiding, and further introduce Byzantine resilience in Private Aggregation for Byzantine-Resilient Heterogeneous Federated Learning by integrating verifiable secret sharing and Private Information Retrieval (PIR).
While computational PIR models allow for features like public verifiability, as noted by Stanislav Kruglik et al. in Two-Server Private Information Retrieval with Optimized Download Rate and Result Verification, information-theoretic variants provide an absolute security ceiling. To make these multi-party protocols practical, researchers focus heavily on reducing communication overhead. Antigoni Polychroniadou et al. address this in Constant-Overhead Unconditionally Secure Multiparty Computation over Binary Fields, utilizing Reverse Multiplication-Friendly Embeddings (RMFE) to achieve constant communication overhead per gate in a malicious, honest-majority setting without sacrificing unconditional security.
Exploiting Physical Channels and Relaxing Perfect Secrecy
When communicating over physical networks, information-theoretic secrecy can also be derived from the physical properties of the transmission medium rather than purely mathematical abstractions. Dong Wang et al. demonstrate in Enhancing Information Security via Physical Layer Approaches in Heterogeneous IoT With Multiple Access Mobile Edge Computing in Smart City how Physical Layer Security (PLS) leverages the inherent randomness, noise, and fading of wireless channels to achieve perfect secrecy. By ensuring the legitimate channel is statistically superior to the eavesdropper's channel, the attacker is physically starved of the information required to decrypt the message. Similarly, Jie Zhang 0076 et al. introduce Multiple Inter-Symbol Obfuscation (MIO) in MIO: Enhancing Wireless Communications Security Through Physical Layer Multiple Inter-Symbol Obfuscation, injecting artificial noise directly into baseband symbols to achieve information-theoretic secrecy without requiring complex multi-antenna hardware.
However, designing practical systems under the strict assumption of a completely unbounded adversary can lead to highly conservative communication rates, a challenge highlighted by Athanasios Papadopoulos et al. in Secret message capacity of a line network and Jianhan Song et al. in Multipath Stealth Communication with Jammers. Furthermore, as Moni Naor et al. prove in Tight Bounds for Unconditional Authentication Protocols in the Manual Channel and Shared Key Models, information-theoretic bounds for authentication often dictate strict entropy requirements that could be relaxed if the adversary faced practical computational limits.
To bridge the gap between the strict requirement of perfect secrecy (where the key must equal the message length) and real-world constraints, researchers have introduced relaxed metrics. Moritz Wiese proposes using mosaics of combinatorial designs in Mosaics of combinatorial designs for information-theoretic security to map messages to specific mathematical structures, minimizing the required shared secret seed.
Taking a different approach, Flavio Calmon et al. introduce $\epsilon$-symbol secrecy and List-Source Codes (LSCs) in Hiding Symbols and Functions: New Metrics and Constructions for Information-Theoretic Security. Instead of trying to hide the entire message perfectly with an insufficient key, their framework spreads uncertainty uniformly across the data. As a result, even an adversary with infinite computing power cannot reliably guess individual symbols or specific functions of the data, demonstrating that meaningful, mathematically absolute security can be achieved even under non-ideal physical constraints.
Key Foundational Work
Some Facets of Complexity Theory and Cryptography: A Five-Lectures Tutorial (2001) — contributed 1 concept elements to this direction.
Zero-Knowledge Arguments and Public-Key Cryptography (1995) — contributed 1 concept elements to this direction.