Rhea : Detecting Privilege-Escalated Evasive Ransomware Attacks Using Format-Aware Validation in the Cloud

Beom Heyn Kim, Seok Min Hong, Mohammad Mannan · 2026

Read Rhea : Detecting Privilege-Escalated Evasive Ransomware Attacks Using Format-Aware Validation in the Cloud by Beom Heyn Kim, Seok Min Hong, Mohammad Mannan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-18216.txt.

Introduction

Beom Heyn Kim, Seok Min Hong, and Mohammad Mannan build a cloud defense against a stealthy class of ransomware. Their 2026 preprint presents Rhea. New ransomware pairs privilege escalation with tricks like intermittent or low-entropy encryption and imitation of safe files. The authors name this class PEER. Such attacks defeat defenses that read input-output patterns, since the attacker can tamper with or blur those traces. Rhea instead validates file formats in a format-aware way.

Checking whether file content truly matches its claimed format is harder to fake. Running the check in the cloud keeps it out of the attacker's reach. The work targets the attacks most likely to beat today's tools.

Catalog record

CREATORBeom Heyn Kim, Seok Min Hong, Mohammad Mannan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-18216

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Rhea : Detecting Privilege-Escalated Evasive Ransomware Attacks Using Format-Aware Validation in the Cloud Beom Heyn Kim, Seok Min Hong, Mohammad Mannan Abstract . Ransomware variants increasingly combine privilege escalation with sophisticated evasion strategies such as intermittent encryption, low-entropy encryption, and imitation attacks. Such powerful ransomware variants, privilege-escalated evasive ransomware (PEER), can defeat existing solutions relying on I/O-pattern analysis by tampering with or obfuscating I/O traces. Meanwhile, conventional statistical content-based detection becomes unreliable as the encryption size decreases due to sampling noises. We present Rhea , a cloud-offloaded ransomware defense system that analyzes replicated data snapshots, so-called mutation snapshots . Rhea introduces Format-Aware Validation that validates the syntactic and semantic correctness of file formats, instead of relying on statistical or entropy-based indicators. By leveraging file-format specifications as detection invariants, Rhea can reliably identify fine-grained and evasive encryption even under elevated attacker privileges. Our evaluation demonstrates that Rhea significantly outperforms existing approaches, establishing its practical effectiveness against modern ransomware threats. Preprint — under review (January 2026) 1. Introduction Ransomware increasingly cause severe operational and financial damage across various domains. In 2024, Ascension, which is one of the largest U.S. healthcare providers, suffered from a ransomware attack leading to the disruption of clinical operations and compromised records of 5.6 million patients and staff ( 54 ) . The rise of ransomware-as-a-service (RaaS) groups such as Medusa has further amplified this trend ( 38 ) . By 2025, Medusa has been used to compromise more than 300 victims, encompassing healthcare, education, insurance, and other critical domains. What makes the situation worse is that the modern ransomware campaigns increasingly rely on privileged execution techniques, such as Bring-Your-Own-Vulnerable-Driver (BYOVD) attacks ( 52 ; 47 ) . Once adversaries gain root privileges, it is not too difficult for attackers to disable security monitors. In parallel, advanced ransomware attacks often employ sophisticated evasion strategies including intermittent encryption ( 49 ; 62 ) , low-entropy encryption, and imitation attacks ( 63 ) . These trends collectively produce a class of Privilege-Escalated Evasive Ransomware (PEER) . Unfortunately, existing ransomware defenses offer limited protection against PEER adversaries. Previous proposals tend to heavily rely on I/O-pattern analysis at the user-, kernel-, hypervisor-, or hardware-level ( 21 ; 46 ; 19 ; 9 ; 2 ; 25 ; 51 ; 5 ; 15 ; 62 ; 37 ; 31 ) . However, I/O-pattern analysis becomes largely untrustworthy under PEER attacks because monitors can be disabled by privileged attackers or neutralized by obfuscated low-level traces. Some systems incorporate content-based checks ( 15 ; 62 ; 37 ; 31 ) . Yet, it is challenging to apply these mechanisms as the sole first-line defense on end-user devices. Content-based detectors partitioned files into chunks and independently analyzed each chunk using statistical methods such as entropy, \chi^{2} statistics, or machine learning. However, these detectors inherit a fundamental statistical limitation: as the inspected window becomes smaller, the empirical byte distribution is overwhelmed by sampling noise. With the sampling noise, the observable statistical features of plaintext, compressed data, and ciphertext rapidly converge and become very difficult to reliably distinguish. Exploiting this inherent limitation of statistical approaches, attackers can evade content-based detection by encrypting only very small spans scattered throughout the file ( 33 ) , which we call fine-grained partial encryption . To overcome the limitation of previous content-based detection, we introduce format-aware validation (FAV), which analyzes not raw byte statistics but the syntactic and semantic invariants of the underlying file format. FAV builds upon a foundational observation that a file format can serve as the correctness specification of the valid data state transition . More specifically, the benign mutation of high-value user files such as text, zip, ooxml, pdf files should always conform to the rich syntactic and semantic structure of the file format. Even a small encrypted span typically violates those invariants, producing malformed text, broken markup, or nonsensical content. Thus, FAV can reliably identify fine-grained partial encryption that is invisible to prior content-based approaches relying on statistical methods. In this work, we develop Rhea , a ransomware defense architecture that offloads both detection and recovery from end devices to the cloud. Cloud offloading offers several advantages: (1) isolation from compromised hosts, including privilege-escalated attacks; (2) abundant storage for long-term retention of data snapshots; (3) ample compute capacity for fine-grained content inspection; and (4) reduced performance and management burden on end devices. Rhea confines PEERs within virtual machines, using VMs as sandboxes to prevent ransomware from escaping the guest environment. Rhea periodically performs a procedure called a scheduled checkpoint , during which it replicates block-level data snapshots—referred to as mutation snapshots —from these VMs to the cloud at designated times (e.g., nightly validation windows or expected idle periods). Rhea then analyzes exclusively the content captured in mutation snapshots using format-aware validation (FAV), rather than relying on ephemeral and potentially untrustworthy I/O streams or coarse statistical content-based heuristics. Implementation and Results. We evaluate a prototype implementation of Rhea using a combination of controlled partial-encryption simulations and format-diverse real-world datasets. Across a wide range of fine-grained encryption patterns—including skip-step and micro-granularity attacks down to single-byte modifications— Rhea achieves nearly 100% detection accuracy for structured file formats such as TXT, PDF, OOXML, and ZIP. In these formats, format-aware validation (FAV) consistently detects encryption-induced violations even when statistical signals are intentionally minimized. Overall, Rhea substantially outperforms entropy- and \chi^{2} -based detectors under fine-grained partial encryption, demonstrating that specification-level correctness is a stronger and more reliable signal than traditional statistical methods. Contributions. To the best of our knowledge, Rhea is the first ransomware detection system that treats file format specifications as security invariants and detects encryption as a violation of those invariants rather than a statistical anomaly. Specifically, this paper makes the following contributions: (1) We identify fundamental limitations of existing ransomware defenses under the PEER threat model, particularly in the presence of fine-grained partial encryption. (2) We present a cloud-offloaded ransomware defense architecture that enables rich content analysis without imposing on-device overhead. (3) We design and implement a novel FAV-based detection pipeline that is resilient to fine-grained partial encryption. (4) We empirically evaluate Rhea against both simulated and real-world ransomware, demonstrating accurate and practical detection. We discuss related work in Section 2 , describe our threat model in Section 3 , present our system design in Section 4 , summarize implementation details in Section 5 , and evaluate our prototype in Section 6 . Section 7 outlines limitations and future work, and Section 8 concludes. 2. Related Work Table 1. Comparison of related approaches to ransomware countermeasures. Each column shows whether the corresponding feature is supported. A ✓ indicates support for the feature, while ✗ denotes the lack of such support. Level Approach Tamper- Resistance Interface Reduction Software-only Solution Cloud-based Detection Format-aware Validation OS Redemption ( 19 ) , ShieldFS ( 9 ) , UNVEIL ( 21 ) , Peeler ( 2 ) , CryptoDrop ( 46 ) , RWGuard ( 27 ) , PayBreak ( 22 ) , CANCAL ( 57 ) , MS-IDS ( 55 ) , ERW-Radar ( 62 ) ✗ ✗ ✓ ✗ ✗ Hardware SSD-Insider ( 5 ) , FlashGuard ( 15 ) , AMOEBA ( 31 ) , RansomBlocker ( 37 ) , LAST (

Read the complete text