Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
Risk-aware Road-network Trajectory Synthesis under Local Differential Privacy The paper presents a framework for road-network trajectory synthesis under local differential privacy to protect user mobility patterns and transition choices. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on road-network trajectory privacy, not AI system deployment safety.
Not All Challenges Are Equal: Evaluating Interactive Liveness Detection Against Real-Time Face Swaps The paper evaluates the effectiveness of randomized interactive gesture challenges in detecting real-time face-swap attacks during remote identity verification. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Evaluates deepfake liveness detection rather than industrial inspection or defect detection.
Not All Challenges Are Equal: Evaluating Interactive Liveness Detection Against Real-Time Face Swaps The paper evaluates the effectiveness of randomized interactive gesture challenges in detecting real-time face-swap attacks during remote identity verification. Preprint	Multimodal models via arxiv · 15 Oct, 06:18	Not relevant Studies interactive face liveness detection instead of general multimodal models.
Revisiting Fast Tendermint: An Agreement Violation and Two Repairs The paper identifies an agreement violation in Fast Tendermint's published round-change rule and proposes two protocol repairs. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses distributed consensus protocol correctness, not AI deployment safety.
Digital-Twin-Referenced Belief-Guided Q-Learning for Jammer Identification and Beam-Power Control in a mmWave Model The paper introduces a digital-twin-referenced belief-guided Q-learning method for jammer identification and beam-power control in millimeter-wave networks. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with RF wireless communications and jammer mitigation, not AI system deployment safety.
Beyond User-Side Defenses: Image Protection in Open-Weight Diffusion Models The paper proposes design requirements and evaluates defenses against model-side protection removal in open-weight image diffusion models. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Focuses on image protection in diffusion models, not industrial inspection.
Beyond User-Side Defenses: Image Protection in Open-Weight Diffusion Models The paper proposes design requirements and evaluates defenses against model-side protection removal in open-weight image diffusion models. Preprint	Multimodal models via arxiv · 15 Oct, 06:18	Not relevant Addresses image generation model security rather than multimodal understanding or reasoning.
Security and robustness analysis of encrypted communications for local home automation: ESPHome Native API and MQTT over TLS 1.3 The paper reviews security and performance trade-offs between MQTT over TLS 1.3 and the ESPHome Native API for local home automation. Peer-reviewed	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant IoT communication security review, not related to AI safety or reliability.
Think Locally Before Computing Jointly: Efficient Secure Multi-Party Computation via Maximizing Local Processing The paper optimizes garbled-circuit secure multi-party computation by restructuring Boolean circuits to maximize local, non-cryptographic evaluations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Cryptographic multi-party computation optimization, not related to AI systems.
The Right to Be Forgotten under Correlation: Privacy-Utility Tradeoffs This paper derives the optimal tradeoff between privacy leakage and utility when redacting information from correlated datasets. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Information-theoretic privacy for data redaction, not AI system safety/reliability.
SEAL-DT: An End-to-End Security Engineering and Assurance Framework for Cross-Domain Digital Twins This paper introduces an end-to-end security engineering framework for specifying, simulating, and assuring cross-domain digital twins. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Framework for digital twin security engineering, not AI safety or reliability.
Can Rényi Accountants Beat Lossless Accountants for DP-SGD? This paper demonstrates that Renyi differential privacy accountants can provide tighter privacy bounds for DP-SGD than lossless accountants. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical differential privacy analysis, not an LLM benchmark.
Can Rényi Accountants Beat Lossless Accountants for DP-SGD? This paper demonstrates that Renyi differential privacy accountants can provide tighter privacy bounds for DP-SGD than lossless accountants. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical DP-SGD privacy bounds, outside deployed AI safety and reliability.
Can Rényi Accountants Beat Lossless Accountants for DP-SGD? This paper demonstrates that Renyi differential privacy accountants can provide tighter privacy bounds for DP-SGD than lossless accountants. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Investigates DP-SGD accounting, not model compression or efficiency.
Tight End-to-End Accounting for Differentially Private Synthetic Data This paper develops accounting techniques to maintain tight differential privacy guarantees across complex synthetic data generation pipelines. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on differential privacy accounting, not small or efficient models.
Tight End-to-End Accounting for Differentially Private Synthetic Data This paper develops accounting techniques to maintain tight differential privacy guarantees across complex synthetic data generation pipelines. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on differential privacy mathematics for synthetic data, not deployed AI safety/reliability.
Tight End-to-End Accounting for Differentially Private Synthetic Data This paper develops accounting techniques to maintain tight differential privacy guarantees across complex synthetic data generation pipelines. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Theoretical privacy accounting framework, not an LLM benchmark.
SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy This paper proposes using state space models to guide black-box image perturbations that disrupt facial recognition systems. Preprint	Computer vision for industrial applications via arxiv · 15 Oct, 06:18	Not relevant Focuses on adversarial face privacy, not industrial inspection or defect detection.
SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy This paper proposes using state space models to guide black-box image perturbations that disrupt facial recognition systems. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Introduces an adversarial attack method rather than an evaluation benchmark.
SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy This paper proposes using state space models to guide black-box image perturbations that disrupt facial recognition systems. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on adversarial attacks against computer vision face recognition, not deployed AI safety/reliability.
SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy This paper proposes using state space models to guide black-box image perturbations that disrupt facial recognition systems. Preprint	Multimodal models via arxiv · 15 Oct, 06:18	Not relevant Deals with face recognition image perturbations, not multimodal reasoning models.
SSM-Attack: Learning Where to Perturb for Black-Box Face Privacy This paper proposes using state space models to guide black-box image perturbations that disrupt facial recognition systems. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on black-box adversarial attacks on face recognition, not model efficiency.
COMPASS: Membership Inference Attack Against Vision Large Language Models via Adaptive Noise Calibration This paper investigates membership inference attacks on vision-language models using low-noise Gaussian perturbations to detect training data exposure. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Does not study model compression, distillation, quantization, or efficient serving.
COMPASS: Membership Inference Attack Against Vision Large Language Models via Adaptive Noise Calibration This paper investigates membership inference attacks on vision-language models using low-noise Gaussian perturbations to detect training data exposure. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on privacy attacks on multimodal models rather than benchmark evaluation.
VulnBump: A Benchmark for Fixing Vulnerable Dependencies in Real Projects While Keeping the Build and Tests Green This paper introduces a benchmark testing whether coding agents can upgrade vulnerable dependencies while preserving project builds and test suites. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Main contribution is an SE benchmark for automated code repair agents.
Fast Forgiving Base64 Decoding with SIMD Instructions This paper introduces a block-oriented SIMD algorithm to accelerate forgiving Base64 decoding while removing whitespace characters. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant SIMD algorithms for Base64 decoding, not related to AI safety.
BrowseSafe++: Benchmarking Prompt Injection Detection Across Web Contexts, Languages, and Browser Views This paper introduces a benchmark of paired web pages to evaluate prompt injection detection across browser agents and views. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on prompt injection evaluation rather than model compression or efficiency.
Extending Layered Attestation into ARM Application Run-Times This paper presents a runtime anchor architecture extending hardware root-of-trust attestation into active application execution on ARM. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on ARM hardware attestation and secure runtimes, not AI.
Local Leakage Resilience of Shamir Secret Sharing: Below Half Threshold Rate This paper provides mathematical bounds proving Shamir secret sharing resilience against local leakage under a lower reconstruction threshold. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Theoretical cryptography on Shamir secret sharing, not AI safety.
Beyond Model Scaling: Rethinking Sequence Construction for Transformer-based Network Intrusion Detection This paper proposes session-coherent sequence temporalization to improve transformer-based network intrusion detection over fixed-window approaches. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on network intrusion detection sequences, not AI safety/reliability.
The Maximum Nonlinearity of 8-variable Balanced Boolean Functions is 116 This paper resolves a mathematical open problem by proving the maximum nonlinearity of 8-variable balanced Boolean functions is 116. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Pure Boolean function mathematics, unrelated to AI safety or reliability.
Privacy-Preserving GNN Inference on Financial Transaction Graphs via Homomorphic Encryption This paper proposes an edge-centric graph neural network inference framework using homomorphic encryption for collaborative multi-bank transaction analysis. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses cryptographic privacy for financial graph networks, not AI safety or guardrails.
Privacy-Preserving GNN Inference on Financial Transaction Graphs via Homomorphic Encryption This paper proposes an edge-centric graph neural network inference framework using homomorphic encryption for collaborative multi-bank transaction analysis. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Presents encrypted GNN inference for AML, not LLM evaluation or benchmarking.
Privacy-Preserving GNN Inference on Financial Transaction Graphs via Homomorphic Encryption This paper proposes an edge-centric graph neural network inference framework using homomorphic encryption for collaborative multi-bank transaction analysis. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on encrypted privacy-preserving inference via CKKS, not small efficient LLMs.
BALLAST: Physics-Aware Moving Target Defense for Industrial Control Systems Using a Digital Twin Safety Oracle This paper develops a moving target network defense that uses digital twins to avoid disrupting industrial control physical processes. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses industrial control system moving target defense, not AI safety or LLM deployment.
HULK: An Efficient Algorithm for Enabling Fast Bulk Bitwise Operations on Homomorphically Encrypted Data This paper introduces an algorithm to speed up bulk bitwise operations on homomorphically encrypted data without increasing ciphertext size. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on homomorphic encryption algorithms, not AI safety or LLM deployment reliability.
Tanchun: Robust Flow-Based HTTP(S) DDoS Detection via Behavioral Attribute Guidance This paper presents a drift-aware system for detecting HTTP DDoS attacks by grouping decryption-free traffic into behavioral sequences. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Covers network-level DDoS traffic detection, not AI safety or deployment reliability.
Kairos: Transactional Security Enforcement at the System Boundary for LLM Agents This paper introduces a transactional isolation monitor that delays irreversible actions to stop indirect prompt injections in software agents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes a system boundary monitor for agents, not an LLM benchmark.
Kairos: Transactional Security Enforcement at the System Boundary for LLM Agents This paper introduces a transactional isolation monitor that delays irreversible actions to stop indirect prompt injections in software agents. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on runtime security enforcement for agents, not model compression.
Towards Safe Autonomy for Computer Use This paper proposes a dual-LLM architecture to establish trust boundaries and prevent indirect prompt injections in visual computer-use agents. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on security boundaries for computer-use agents, not model efficiency.
Towards Safe Autonomy for Computer Use This paper proposes a dual-LLM architecture to establish trust boundaries and prevent indirect prompt injections in visual computer-use agents. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes a security architecture against prompt injection, not an LLM benchmark.
Prompt-Segmentation Meets ROI Encryption: Fine-Grained Video Protection Based on Diffusion Isolation This paper presents a prompt-segmentation method to encrypt irregular video regions with less distortion in surrounding non-sensitive areas. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on video codec ROI encryption, not AI safety, alignment, or LLM guardrails.
X.js: A Live Benchmark for Node.js Taint-Style Vulnerabilities The paper presents X.js, a benchmark of real-world Node.js taint-style vulnerabilities paired with executable exploits and patches. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on web application taint vulnerabilities, not AI safety or deployment reliability.
X.js: A Live Benchmark for Node.js Taint-Style Vulnerabilities The paper presents X.js, a benchmark of real-world Node.js taint-style vulnerabilities paired with executable exploits and patches. Preprint	LLMs for code via arxiv · 15 Oct, 06:18	Not relevant Introduces a Node.js security vulnerability dataset, not LLMs for code or software agents.
Autonomous Cyber Operations Gym for Training and Evaluating AI Models for Defence The paper introduces a simulation gym for training and evaluating autonomous cyber defence models and network intrusion detectors. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Proposes an emulation gym for cyber operations, not an LLM evaluation benchmark.
Autonomous Cyber Operations Gym for Training and Evaluating AI Models for Defence The paper introduces a simulation gym for training and evaluating autonomous cyber defence models and network intrusion detectors. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Focuses on cyber network emulation, not model compression or efficient serving.
Autonomous Cyber Operations Gym for Training and Evaluating AI Models for Defence The paper introduces a simulation gym for training and evaluating autonomous cyber defence models and network intrusion detectors. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Trains intrusion detection classifiers, not safety or reliability of LLM deployments.
ARACDE: Retrofitting Databases with ABAC using External Authorization Engines The paper presents ARCADE, an architecture decoupling relational database authorization logic into external attribute-based policy engines. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Addresses relational database access control, not AI safety or LLM deployment reliability.
Dynamic Attack Surfaces: A Queue of Branching Processes The paper models dynamic software attack surfaces and patching queues using branching processes under evolving attacker populations. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Models cybersecurity vulnerability patching dynamics, not AI deployment safety or guardrails.
$\mathsf{EFI}\Leftrightarrow\mathsf{1PRS}$ The paper proves an equivalence between EFI pairs and expanding single-copy pseudorandom state generators in quantum cryptography. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on theoretical quantum cryptography, not AI safety or deployment robustness.
AegisAsync: Symbolic Async-Event Simulation for Risks in Python Asynchronous Programs The paper introduces AegisAsync, a symbolic event simulation framework for detecting concurrency and cancellation bugs in asynchronous Python code. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Focuses on static analysis of Python async code, not AI safety or LLM deployment risks.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	AI agents via arxiv · 15 Oct, 06:18	Not relevant Presents competition pipelines for financial NLP tasks, not autonomous multi-step agents.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	Small and efficient models via arxiv · 15 Oct, 06:18	Not relevant Applies existing 32B models without addressing model compression or inference efficiency.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant A shared-task competition entry, not an evaluation methodology or benchmark study.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	LLMs for code via arxiv · 15 Oct, 06:18	Not relevant Focuses on financial text tasks like NER and QA, not software engineering or code.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	Multimodal models via arxiv · 15 Oct, 06:18	Not relevant Processes text-only financial documents, not multimodal data.
Fin_Agent at FinEval 2026: Task-Aware Methods for Multilingual Financial NLP The paper describes a system submission combining rules and language models for multilingual financial named-entity recognition and question answering. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Applies NLP models to financial tasks, not AI safety or deployment guardrails.
Does System One Transfer? Calibrated Typed-Decision Models, Label Budgets, and Translation for Indonesian Regional Languages The paper evaluates the cross-lingual transfer, calibration, and translation sensitivity of fast typed-decision models on regional Indonesian languages. Preprint	LLM evaluation and benchmarks via arxiv · 15 Oct, 06:18	Not relevant Focuses on cross-lingual evaluation of classifiers rather than general LLM benchmark analysis.
Does System One Transfer? Calibrated Typed-Decision Models, Label Budgets, and Translation for Indonesian Regional Languages The paper evaluates the cross-lingual transfer, calibration, and translation sensitivity of fast typed-decision models on regional Indonesian languages. Preprint	AI safety and reliability in deployment via arxiv · 15 Oct, 06:18	Not relevant Deals with cost reduction and classification accuracy, not safety or guardrails.
Does System One Transfer? Calibrated Typed-Decision Models, Label Budgets, and Translation for Indonesian Regional Languages The paper evaluates the cross-lingual transfer, calibration, and translation sensitivity of fast typed-decision models on regional Indonesian languages. Preprint	Retrieval-augmented generation (RAG) via arxiv · 15 Oct, 06:18	Not relevant Studies text classification and confidence routing, not retrieval-augmented generation.

10,321–10,380 of 17,313← Previous Next →