Selected work

Representative Papers

Are Language Models Script-Aware?

Oct 06, 2026

This study addresses a critical gap in language model research, which has predominantly focused on language selection while overlooking fundamental script-level knowledge, particularly regarding generation in non-target scripts. To bridge this gap, this work shifts the focus toward script recognition and adaptation capabilities by constructing a multilingual test set spanning multiple writing systems. Two complementary experimental paradigms—input adaptation and explicit instruction following—are designed to comparatively evaluate the script-processing proficiency of models across varying scales. The findings demonstrate that these models possess substantial script knowledge, achieving over 98% fidelity in Latin scripts. Moreover, larger models significantly outperform smaller counterparts under non-standard script combinations. By systematically investigating graphical symbol recognition capabilities, this research fills a notable void in the existing literature on the orthographic competencies of large language models.

0 citationsRead paper

Script Choice in LLMs: Evidence for Late-Layer Commitment

Sep 23, 2026

This study investigates the inter-layer distribution mechanisms of script knowledge and the formation of output script commitment in large language models. Using logistic regression probes and LogitLens analysis, we reveal a two-stage asymmetry in script processing: input and instruction scripts are encoded in early layers, intermediate layers default to Latin characters, and actual output script commitment emerges exclusively in the final layer. Our findings demonstrate that this commitment capacity correlates positively with model depth, while smaller models exhibit comparatively weaker performance. These results establish the critical role of model depth in multilingual generation, offering new perspectives for designing deeper and more inclusive multilingual architectures.

0 citationsRead paper

Chi-MERA: Defending Orbit-Based Authentication of LEO Satellites with the Space Oddity of MLAT (Long Version)

Jul 21, 2026

This work addresses the vulnerability of physical-layer authentication in low Earth orbit (LEO) satellite systems to coordinated multi-device spoofing attacks, which can induce false alarm rates as high as 40%. To counter this threat, the authors propose Chi-MERA, a novel scheme that integrates multilateration (MLAT) residual analysis with a resilient signature mechanism that eliminates the need for a single trusted reference receiver. By leveraging orbital dynamics to distinguish between genuine satellites and terrestrial spoofers, Chi-MERA significantly enhances authentication robustness. In a network of n receivers, the method tolerates up to n/2 compromised devices while achieving a false alarm rate below 2% and a miss rate of approximately 3%; under extreme conditions, the false alarm rate further drops below 1%.

0 citationsRead paper

From Model to Breach: Towards Actionable LLM-Generated Vulnerabilities Reporting

Nov 06, 2025

This paper addresses the insufficient risk assessment of security vulnerabilities introduced by LLM-based programming assistants. We propose the first risk-aware evaluation framework integrating vulnerability severity, generation probability, and prompt exposure (PE)—a novel metric quantifying the susceptibility of vulnerabilities to adversarial prompting. We further introduce model exposure (ME) to measure vulnerability prevalence across models. Empirical analysis reveals that even for long-disclosed vulnerabilities, mainstream open-source code-generation models remain significantly susceptible, confirming a fundamental trade-off between security and functionality. Our contributions are threefold: (1) formal definition and empirical validation of the PE/ME dual-metric framework; (2) establishment of an actionable vulnerability prioritization mechanism grounded in quantitative risk estimation; and (3) identification of critical limitations in current security hardening techniques under realistic prompt distributions—thereby providing both theoretical foundations and practical guidance for targeted remediation of high-risk vulnerabilities.

0 citationsRead paper

The complexity of the SupportMinors Modeling for the MinRank Problem

Jun 06, 2025

The MinRank problem underpins several post-quantum cryptographic schemes (e.g., GeMSS, Rainbow), yet the algebraic SupportMinors modeling approach lacked a rigorous asymptotic complexity analysis. Method: We integrate tools from algebraic geometry, Gröbner basis theory, and polynomial modeling of rank constraints to conduct a precise asymptotic analysis of the SupportMinors system. Contribution/Results: We establish the first tight asymptotic bounds—both upper and lower—on the solving complexity of SupportMinors. Our analysis formally validates the original heuristic complexity estimates while correcting several implicit assumptions, thereby filling a long-standing theoretical gap in the complexity assessment of this algebraic attack. The results yield the first verifiable security benchmark for MinRank-based cryptosystems, significantly enhancing the rigor and practical applicability of algebraic cryptanalysis.

0 citationsRead paper

Recent publications

Latest Papers

Are Language Models Script-Aware?

Oct 06, 2026

This study addresses a critical gap in language model research, which has predominantly focused on language selection while overlooking fundamental script-level knowledge, particularly regarding generation in non-target scripts. To bridge this gap, this work shifts the focus toward script recognition and adaptation capabilities by constructing a multilingual test set spanning multiple writing systems. Two complementary experimental paradigms—input adaptation and explicit instruction following—are designed to comparatively evaluate the script-processing proficiency of models across varying scales. The findings demonstrate that these models possess substantial script knowledge, achieving over 98% fidelity in Latin scripts. Moreover, larger models significantly outperform smaller counterparts under non-standard script combinations. By systematically investigating graphical symbol recognition capabilities, this research fills a notable void in the existing literature on the orthographic competencies of large language models.

0 citationsRead paper

Script Choice in LLMs: Evidence for Late-Layer Commitment

Sep 23, 2026

This study investigates the inter-layer distribution mechanisms of script knowledge and the formation of output script commitment in large language models. Using logistic regression probes and LogitLens analysis, we reveal a two-stage asymmetry in script processing: input and instruction scripts are encoded in early layers, intermediate layers default to Latin characters, and actual output script commitment emerges exclusively in the final layer. Our findings demonstrate that this commitment capacity correlates positively with model depth, while smaller models exhibit comparatively weaker performance. These results establish the critical role of model depth in multilingual generation, offering new perspectives for designing deeper and more inclusive multilingual architectures.

0 citationsRead paper

Chi-MERA: Defending Orbit-Based Authentication of LEO Satellites with the Space Oddity of MLAT (Long Version)

Jul 21, 2026

This work addresses the vulnerability of physical-layer authentication in low Earth orbit (LEO) satellite systems to coordinated multi-device spoofing attacks, which can induce false alarm rates as high as 40%. To counter this threat, the authors propose Chi-MERA, a novel scheme that integrates multilateration (MLAT) residual analysis with a resilient signature mechanism that eliminates the need for a single trusted reference receiver. By leveraging orbital dynamics to distinguish between genuine satellites and terrestrial spoofers, Chi-MERA significantly enhances authentication robustness. In a network of n receivers, the method tolerates up to n/2 compromised devices while achieving a false alarm rate below 2% and a miss rate of approximately 3%; under extreme conditions, the false alarm rate further drops below 1%.

0 citationsRead paper

From Model to Breach: Towards Actionable LLM-Generated Vulnerabilities Reporting

Nov 06, 2025

This paper addresses the insufficient risk assessment of security vulnerabilities introduced by LLM-based programming assistants. We propose the first risk-aware evaluation framework integrating vulnerability severity, generation probability, and prompt exposure (PE)—a novel metric quantifying the susceptibility of vulnerabilities to adversarial prompting. We further introduce model exposure (ME) to measure vulnerability prevalence across models. Empirical analysis reveals that even for long-disclosed vulnerabilities, mainstream open-source code-generation models remain significantly susceptible, confirming a fundamental trade-off between security and functionality. Our contributions are threefold: (1) formal definition and empirical validation of the PE/ME dual-metric framework; (2) establishment of an actionable vulnerability prioritization mechanism grounded in quantitative risk estimation; and (3) identification of critical limitations in current security hardening techniques under realistic prompt distributions—thereby providing both theoretical foundations and practical guidance for targeted remediation of high-risk vulnerabilities.

0 citationsRead paper

The complexity of the SupportMinors Modeling for the MinRank Problem

Jun 06, 2025

The MinRank problem underpins several post-quantum cryptographic schemes (e.g., GeMSS, Rainbow), yet the algebraic SupportMinors modeling approach lacked a rigorous asymptotic complexity analysis. Method: We integrate tools from algebraic geometry, Gröbner basis theory, and polynomial modeling of rank constraints to conduct a precise asymptotic analysis of the SupportMinors system. Contribution/Results: We establish the first tight asymptotic bounds—both upper and lower—on the solving complexity of SupportMinors. Our analysis formally validates the original heuristic complexity estimates while correcting several implicit assumptions, thereby filling a long-standing theoretical gap in the complexity assessment of this algebraic attack. The results yield the first verifiable security benchmark for MinRank-based cryptosystems, significantly enhancing the rigor and practical applicability of algebraic cryptanalysis.

0 citationsRead paper