CCS 2026: Decoupled TFHE Bootstrapping & LLM-Driven Web Scanning Find 59 Zero-Days
The article summarizes two CCS 2026 papers: a decoupled TFHE functional bootstrapping scheme that separates ciphertext conversion from LUT evaluation to boost efficiency, and SemScanner, an LLM-powered semantic-aware black-box scanner that uncovered 59 zero-day vulnerabilities across 20 real web applications by reasoning about page functionality and user interactions.
Paper 1: Decoupled TFHE-type Functional Bootstrapping (DFBS)
Functional Bootstrapping (FBS) in TFHE refreshes ciphertexts while evaluating a lookup table (LUT). In existing constructions the target LUT is embedded in the accumulator’s initial state and evaluated via blind rotation. This coupling limits the efficiency and functional extensibility of several bootstrapping variants.
The authors propose Decoupled TFHE-type Functional Bootstrapping (DFBS) , splitting FBS into two independent phases:
Ciphertext conversion : a single blind rotation converts an LWE ciphertext into an RGSW ciphertext.
External product : the external product then performs denoising and function evaluation.
Because the conversion result is independent of the LUT, it can be reused across multiple target functions. Building on this separation, the paper improves several TFHE bootstrapping variants:
Full-domain FBS : a construction requiring only one blind rotation, further extended to support multi-value bootstrapping and encrypted LUTs.
Multi-input FBS : two tree-based methods are introduced. The external product tree reduces the number of blind rotations from exponential growth to linear growth, and can be transformed into a CMux tree to efficiently handle functions with larger input sizes.
High-precision FBS : by integrating homomorphic digit decomposition with multi-input FBS, redundant blind rotations in multi-input bootstrapping are eliminated, lowering computational overhead.
Paper 2: SemScanner – Semantic-Aware Black-Box Scanning for Hidden Web Attack Surfaces
Modern web applications heavily use dynamic client-side rendering. Many HTTP requests are triggered only after specific user interactions, creating a hidden attack surface that traditional black-box scanners miss because they lack semantic understanding of the application.
SemScanner addresses this by leveraging a large language model (LLM) to:
Understand page functionality.
Infer user interaction tasks.
Execute those tasks via an autonomous agent to discover hidden requests.
During scanning, SemScanner builds a web application semantic graph that links page functions, interaction tasks, requests, and their contexts. This graph guides a vulnerability detection agent to perform targeted testing.
Evaluation on 20 real-world web applications yielded 59 previously unknown zero-day vulnerabilities – nearly three times the count found by the strongest baseline. All vulnerabilities were responsibly disclosed and assigned CVE identifiers.
Event Context
The article announces a live session (November 12, 2026, 18:00–19:00 CST) on WeChat Video and Bilibili, featuring the two paper authors:
Zhihao Li – Assistant Researcher, Ant Group Technology Research Institute.
Fukun Mei – PhD Student, Tsinghua University.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
