In ideal cryptography, an encrypted message is like a sealed black box: you cannot see what is inside, and you certainly cannot change what is inside without breaking the box. However, many foundational public-key primitives, such as raw RSA or ElGamal, are "malleable." They are more like clay than a black box—an attacker can reshape the ciphertext into a new, valid ciphertext that decrypts to a mathematically related message, even if they cannot read the original.
The research direction across these papers focuses on the tension between this inherent malleability and the "gold standard" of modern security: Chosen-Ciphertext Attack (CCA) security.
The Malleability Trap
To understand why malleability is dangerous, consider the "Decryption Oracle" scenario. In a security game, an attacker is given a challenge ciphertext $C$ (the encryption of a secret $M$). The attacker is allowed to ask a server to decrypt any ciphertext except $C$.
If the encryption scheme is malleable, the attacker can mathematically transform $C$ into $C'$. For example, in raw RSA, multiplying a ciphertext by $2^e$ results in a ciphertext that decrypts to $2M$. As Hyung Tae Lee and San Ling demonstrate in their analysis of homomorphic schemes, an attacker can "re-randomize" a ciphertext to create a new version that looks different but contains the same underlying data. They then submit this modified $C'$ to the server, receive the result, and easily compute the original secret $M$.
Rigid Padding as a Defense
The standard way to kill malleability is through padding. By adding a specific, verifiable structure to the plaintext before encrypting it, we ensure that any random "tweak" to the ciphertext will result in a bitstring that fails the padding check upon decryption.
Dan Boneh explores this in Simplified OAEP for the RSA and Rabin Functions. While the industry standard OAEP (Optimal Asymmetric Encryption Padding) uses a complex two-round Feistel network to ensure non-malleability, Boneh shows that for specific functions like RSA, a simpler one-round scheme called SAEP is sufficient. The intuition is that the padding acts as a "binding" mechanism: it forces the mathematical structure of the function to collide with the rigid requirements of the padding, making it computationally impossible for an attacker to create a related ciphertext that still "looks" valid to the decryptor.
The Implementation Gap
Even when the theory is sound, the implementation often fails to enforce this rigidity. This is a recurring theme in applied security research:
- Incomplete Authentication: Claudio Orlandi and Jakob Jakobsen found that Telegram's MTProto protocol failed to include padding in its integrity checks. This allowed attackers to append random data to the end of a ciphertext. If the decryption process ignored this "trailing" data, the modified ciphertext remained valid, violating the principle of non-malleability.
- Middleware Neglect: In the emerging Model Context Protocol (MCP) for AI, Biwei Yan and colleagues uncovered widespread "cryptographic misuse". Developers often use encryption modes that provide confidentiality but lack integrity protection (MACs), leaving the system vulnerable to tampering.
- HPC Vulnerabilities: Abu Naser and others noted that high-performance computing libraries often prioritize speed over authenticated encryption, leaving data in transit susceptible to active tampering.
The Fundamental Conflict
Perhaps the most intriguing aspect of this research is the "Homomorphic Paradox." Homomorphic encryption is designed to be malleable—it allows you to perform addition or multiplication on encrypted data.
However, as Lee and Ling argue, this useful malleability is fundamentally at odds with CCA2 security. If a user can legitimately transform a ciphertext to perform a calculation, an attacker can use that same transformation to "probe" a decryption oracle. Their work proves that many schemes claiming to achieve both are mathematically unsound. You can have a system that is flexible (homomorphic) or a system that is rigid (CCA2 secure), but bridging the two requires complex "binding" techniques that are difficult to get right.
In summary, the research shows that security is not just about keeping secrets; it is about ensuring that ciphertexts are immutable. Whether through simplified padding like SAEP or more robust authenticated encryption, the goal is to ensure that any modification to a ciphertext—no matter how clever the math—results in nothing but an error.