arXiv subjects

Jürgen Cito

Publications and source records attributed to Jürgen Cito.

At least 19 recordsLinked to original sources

arXiv2026

Recognition Without Mitigation: Ethical Frameworks in Autonomous Offensive-LLM Agent Research

Large language models have moved from advising on offensive security to autonomously conducting it. A growing literature presents agents that execute reconnaissance, exploitation, and privilege escalation against real or simulated targets. Such an agent is a deployable, re-pointable capability that could be used by a malicious actor against a non-consenting third party. Papers that introduce these prototypes therefore carry an ethical burden, which top security venues have begun to encode as hard policy in their 2026 ethics mandates. We present a systematic audit of ethics reporting based on 54 papers describing autonomous offensive-LLM penetration-testing prototypes (2023-2026), assembled from peer-reviewed venues as well as from pre-prints. We score each against an eleven-dimension instrument derived both top-down from the Menlo Report, and bottom-up from 2026 security venue ethics mandates. Our central result is a recognition-without-mitigation gap: dual-use risk is reported as recognized in 57% of papers, but a concrete mitigation is reported in only 15%, roughly a 4:1 gap. Safeguards commonly protect the experiment, not the public. Guardrail bypasses are deployed by 17% of papers but not disclosed to LLM providers. Measured against the new mandates, the corpus defines a pre-regulation baseline in which current practice does not meet the substantive requirements. We argue this audit is itself defensive intelligence on the offensive-agent ecosystem, and provide an ethics statement checklist for authors working on autonomous offensive agents.

cs.CR↗

arXiv2026

Lost in Code Generation: Reimagining the Role of Software Models in AI-driven Software Engineering

Generative AI enables rapid "vibe coding" and agentic software engineering, where natural-language prompts yield working software systems. This lowers barriers to software creation, but it also collapses the boundary between prototypes and engineered software. Systems that appear complete may lack robustness, security, and maintainability. We argue that this shift motivates a renewed role for software models. Rather than serving only as upfront blueprints, models can be recovered from AI-generated systems, used to restore comprehension, and refined to guide subsequent evolution. We distinguish an agentic loop, in which recovered models support automated checking and repair, from a human reflective loop, in which models expose assumptions for inspection and revision. We identify candidate model types and illustrate how recovered models can expose implicit assumptions as possible constraints for review and validation. This paper positions software models as a bridge between current software engineering practice and model-based reasoning in AI-driven development.

cs.SE↗

arXiv2026

ATLAS: Discovering Agent Strategies through LLM-Guided Abstraction and Automata Learning

Large Language Model (LLM)-based agents are increasingly used for complex tasks such as software testing and cybersecurity assessment. While these agents demonstrate impressive capabilities, their behavior is difficult to understand, explain, and analyze. Existing evaluations focus mainly on task success and execution traces, offering limited insight into the strategies employed by the agent. We present ATLAS (Automata Learning for Agent Trajectory Analysis and Strategy Discovery), an approach for recovering interpretable behavioral models from agent trajectories. ATLAS combines trace abstraction with automata learning to infer finite-state models that capture observed agent-environment interaction strategies. These models provide human-interpretable insights and support automated analyses of recurring behaviors, decision points, successful task-completion paths, and failure loops. As a proof of concept, we apply ATLAS to trajectories generated by an LLM-based penetration-testing agent. The resulting models expose high-level behavioral strategies for exploiting vulnerable machines that are difficult to identify from raw execution traces alone. We discuss how learned behavioral models can support explainability, model-guided exploration, auditing, and analysis of agentic systems. We further demonstrate symbolic model-based knowledge transfer from powerful frontier models to compact language models. In addition, we show how model transformations can derive concise explanations of agent behavior in a penetration-testing case study comprising 12 vulnerable machines. ATLAS highlights a new opportunity for model-driven engineering: transforming agent trajectories into explicit behavioral models that enable systematic understanding and analysis of otherwise opaque AI agents.

cs.SE↗

arXiv2026

Enhancing Linux Privilege Escalation Attack Capabilities of Local LLM Agents

Cloud-based Large Language Models (LLMs) can perform autonomous penetration-testing sub-tasks such as Linux privilege escalation, but raise security, privacy, and sovereignty concerns. Locally hosted open-weight models avoid these issues, yet prior work reports that small open-weight models succeed on only 8-16% of standardized privilege-escalation tasks, far below frontier cloud models. This paper is an empirical study of why small models fail at this task and which engineering techniques close the gap. From execution traces we distill six recurring failure modes, map each to an established enhancement technique, and evaluate five (chain-of-thought prompting, retrieval-augmented generation, structured prompting, history compression, and reflective analysis) as reproducible extensions to the open-source hackingBuddyGPT framework. Under a single shared harness and matched conditions, the set of techniques we evaluate raise two SLMs (Llama3.1 8B, Qwen2.5 7B) from 8% to 67% with guidance, matching guided GPT-4o. A larger open-weight reference model (Llama3.1 70B) reaches 83%. A full-factorial ablation shows that reflection-based techniques contribute most and reveals vulnerability discovery, not exploitation, as the main constraint for local models. We report these as transferable lessons for building reliable local offensive agents as well as to inform defenders.

cs.CR↗

arXiv2026

The Ethics of Autonomous AI Agents for Offensive Security

LLM-driven autonomous agents are reshaping offensive security. Unlike traditional penetration-testing tooling - deterministic, narrowly scoped, and operated by trained practitioners - agentic security tools exhibit indeterminacy along three independent dimensions. First, their actions are drawn from a non-deterministic policy whose outputs resist both ex-ante and ex-post explanation. This complicates incident attribution and pre-deployment safety reviews. Second, their impact is open-ended due to their non-deterministic actions, agency of utilized models, and opaque LLM supply-chains. Third, their user population is indeterminate in both size and required skill: the operating skill floor for using or developing offensive capabilities has dropped sharply. These three properties are linked thematically, but are not derivable from one another. Combined with the structural cost asymmetry between offense and defense, they enable the industrialization of offensive capability. The net short-term effect favors attackers, even if the same technology may, in the long run, democratize access to defensive practice. Existing dual-use cybersecurity and AI-ethics frameworks struggle to address this combination. Our work analyzes how moral attribution becomes diffuse between users, tool-makers, and third parties when employing autonomous AI agents for offensive security. We also examine the stakeholder impact of this technology and provide stratified recommendations.

cs.CR↗

arXiv2026

Cochise: A Reference Harness for Autonomous Penetration Testing

Recent work on LLM-driven autonomous penetration testing reports promising results, but existing systems often bundle architectural, prompting, and tool-integration choices together. This makes it difficult to determine what is gained over a simple agent and harness. We present Cochise, a 630 LOC Python reference implementation for autonomous penetration-testing experiments. Cochise connects to a Linux execution host over SSH and supports attacking controlled target environments reachable from that jump host. The prototype implements a Planner--Executor architecture in which long-term state is maintained by the planner, while a ReAct-style executor issues commands over SSH and self-corrects based on command outputs. The scenario prompt can be adapted to different target environments. We evaluate the harness against a live third-party testbed, Game of Active Directory (GOAD). Cochise is intended not as a state-of-the-art penetration-testing agent, but as a reusable experimental infrastructure for comparing models, agent architectures, and penetration-testing traces. Alongside the prototype, we release replay and analysis tools: (i) cochise-replay for offline visualization of captured runs, (ii) cochise-analyze-logs and cochise-analyze-graphs for cost, token, duration, and compromise analysis, and (iii) a corpus of JSON trajectory logs from GOAD runs, so that researchers can study agent behavior without provisioning the 48--64 GB RAM / 190 GB storage testbed themselves. Tool demo video available at https://youtu.be/2mQimB1ufyI.

cs.CR↗

arXiv2026

Static Factorisation of Probabilistic Programs With User-Labelled Sample Statements and While Loops

It is commonly known that any Bayesian network can be implemented as a probabilistic program, but the reverse direction is not so clear. In this work, we address the open question to what extent a probabilistic program with user-labelled sample statements and while loops - features found in languages like Gen, Turing, and Pyro - can be represented graphically. To this end, we extend existing operational semantics to support these language features. By translating a program to its control-flow graph, we define a sound static analysis that approximates the dependency structure of the random variables in the program. As a result, we obtain a static factorisation of the implicitly defined program density, which is equivalent to the known Bayesian network factorisation for programs without loops and constant labels, but constitutes a novel graphical representation for programs that define an unbounded number of random variables via loops or dynamic labels. We further develop a sound program slicing technique to leverage this structure to statically enable three well-known optimisations for the considered program class: we reduce the variance of gradient estimates in variational inference and we speed up both single-site Metropolis Hastings and sequential Monte Carlo. These optimisations are proven correct and empirically shown to match or outperform existing techniques.

cs.PL↗

arXiv2026

Towards Reliable Local Security Agents: Verifiable Post-Training for Linux Privilege Escalation

LLM agents are becoming increasingly important in the security domain, but leading systems are often closed-source, cloud-based, hard to reproduce or use with sensitive code. This creates a need for small, local models that can perform security tasks under strict resource constraints, though effective methods for developing them remain unexplored. In this paper, we address this gap by proposing a two-stage post-training recipe that turns a small local language model into a security agent. To this end, we focus on Linux privilege escalation as a representative setting to systematically study the training of local models, as the task is both automatically verifiable and requires multi-step interactive reasoning. Using an experimental setup that mitigates data leakage, we post-train a small 4B model in two stages: supervised fine-tuning on traces from procedural privilege-escalation environments, followed by reinforcement learning with verifiable rewards. On a held-out benchmark of 12 Linux privilege-escalation scenarios, supervised fine-tuning doubles the baseline success rate under a tight budget of 20 interaction rounds, and subsequent reinforcement learning training improves our model, PrivEsc-LLM 4B, to 93.3% success, behind only Claude Opus 4.7 at this budget. At the same time, the expected inference cost per successful escalation decreases by more than 80x. Our findings not only show that small local models can be adapted to complex security tasks, but also document the challenges involved, offering guidance for transferring this recipe to other settings.

cs.CR↗

arXiv2026

Reducing Token Usage of State-in-Context Agents using Minification

This paper presents a replication and extension of the recently introduced state-in-context agent framework. We independently re-implement the DirectSolve variant and evaluate it on the SWE-bench Verified benchmark. We report end-to-end full-benchmark results using GPT-5-mini and run selected ablations with GPT-4.1. In addition, we investigate a complementary research question: What is the impact of token-reducing input transformation strategies on the performance of software engineering agents? Based on a preliminary prompt analysis, we identify source code as the dominant contributor to token consumption. We therefore apply a series of code minification techniques that remove or shorten non-essential lexical elements while preserving program semantics. The proposed transformations are integrated into the agent and systematically evaluated. Experiments show that minification reduces average input token usage by 42% with a 12 percentage-point drop in resolution rate. These findings demonstrate that lightweight source code transformations can yield substantial efficiency gains while retaining a substantial fraction of the baseline performance, indicating a promising path toward more cost-effective agents. The full implementation is publicly available on GitHub: https://github.com/ipa-lab/minified-state-in-context-agent

cs.SE↗

arXiv2026

Can LLMs Hack Enterprise Networks? -- Replicated Computational Results (RCR) Report

This is the Replicated Computational Results (RCR) Report for the paper ``Can LLMs Hack Enterprise Networks?" The paper empirically investigates the efficacy and effectiveness of different LLMs for penetration-testing enterprise networks, i.e., Microsoft Active Directory Assumed-Breach Simulations. This RCR report describes the artifacts used in the paper, how to create an evaluation setup, and highlights the analysis scripts provided within our prototype.

cs.CR↗

arXiv2026

Adversarial Bug Reports as a Security Risk in Language Model-Based Automated Program Repair

Large Language Model (LLM) - based Automated Program Repair (APR) systems are increasingly integrated into modern software development workflows, offering automated patches in response to natural language bug reports. However, this reliance on untrusted user input introduces a novel and underexplored attack surface. In this paper, we investigate the security risks posed by adversarial bug reports -- realistic-looking issue submissions crafted to mislead APR systems into producing insecure or harmful code changes. We develop a comprehensive threat model and conduct an empirical study to evaluate the vulnerability of APR systems to such attacks. Our demonstration comprises 51 adversarial bug reports generated across a spectrum of strategies, ranging from manual curation to fully automated pipelines. We test these against a leading LLM-based APR system and assess both pre-repair defenses (e.g., LlamaGuard variants, PromptGuard variants, Granite-Guardian, and custom LLM filters) and post-repair detectors (GitHub Copilot, CodeQL). Our findings show that current defenses are insufficient: 90% of crafted bug reports triggered attacker-aligned patches. The best pre-repair filter blocked only 47%, while post-repair analysis -- often requiring human oversight -- was effective in just 58% of cases. To support scalable security testing, we introduce a prototype framework for automating the generation of adversarial bug reports. Our analysis exposes a structural asymmetry: generating adversarial inputs is inexpensive, while detecting or mitigating them remains costly and error-prone. We conclude with recommendations for improving the robustness of APR systems against adversarial misuse and highlight directions for future work on secure APR.

cs.SE↗

arXiv2025

Online and Interactive Bayesian Inference Debugging

Probabilistic programming is a rapidly developing programming paradigm which enables the formulation of Bayesian models as programs and the automation of posterior inference. It facilitates the development of models and conducting Bayesian inference, which makes these techniques available to practitioners from multiple fields. Nevertheless, probabilistic programming is notoriously difficult as identifying and repairing issues with inference requires a lot of time and deep knowledge. Through this work, we introduce a novel approach to debugging Bayesian inference that reduces time and required knowledge significantly. We discuss several requirements a Bayesian inference debugging framework has to fulfill, and propose a new tool that meets these key requirements directly within the development environment. We evaluate our results in a study with 18 experienced participants and show that our approach to online and interactive debugging of Bayesian inference significantly reduces time and difficulty on inference debugging tasks.

cs.SE↗

arXiv2025

Can LLMs Hack Enterprise Networks? Autonomous Assumed Breach Penetration-Testing Active Directory Networks

Enterprise penetration-testing is often limited by high operational costs and the scarcity of human expertise. This paper investigates the feasibility and effectiveness of using Large Language Model (LLM)-driven autonomous systems to address these challenges in real-world Active Directory (AD) enterprise networks. We introduce a novel prototype designed to employ LLMs to autonomously perform Assumed Breach penetration-testing against enterprise networks. Our system represents the first demonstration of a fully autonomous, LLM-driven framework capable of compromising accounts within a real-life Microsoft Active Directory testbed, GOAD. We perform our empirical evaluation using five LLMs, comparing reasoning to non-reasoning models as well as including open-weight models. Through quantitative and qualitative analysis, incorporating insights from cybersecurity experts, we demonstrate that autonomous LLMs can effectively conduct Assumed Breach simulations. Key findings highlight their ability to dynamically adapt attack strategies, perform inter-context attacks (e.g., web-app audits, social engineering, and unstructured data analysis for credentials), and generate scenario-specific attack parameters like realistic password candidates. The prototype exhibits robust self-correction mechanisms, installing missing tools and rectifying invalid command generations. We find that the associated costs are competitive with, and often significantly lower than, those incurred by professional human pen-testers, suggesting a path toward democratizing access to essential security testing for organizations with budgetary constraints. However, our research also illuminates existing limitations, including instances of LLM ``going down rabbit holes'', challenges in comprehensive information transfer between planning and execution modules, and critical safety concerns that necessitate human oversight.

cs.CR↗

arXiv2025

On the Surprising Efficacy of LLMs for Penetration-Testing

This paper presents a critical examination of the surprising efficacy of Large Language Models (LLMs) in penetration testing. The paper thoroughly reviews the evolution of LLMs and their rapidly expanding capabilities which render them increasingly suitable for complex penetration testing operations. It systematically details the historical adoption of LLMs in both academic research and industry, showcasing their application across various offensive security tasks and covering broader phases of the cyber kill chain. Crucially, the analysis also extends to the observed adoption of LLMs by malicious actors, underscoring the inherent dual-use challenge of this technology within the security landscape. The unexpected effectiveness of LLMs in this context is elucidated by several key factors: the strong alignment between penetration testing's reliance on pattern-matching and LLMs' core strengths, their inherent capacity to manage uncertainty in dynamic environments, and cost-effective access to competent pre-trained models through LLM providers. The current landscape of LLM-aided penetration testing is categorized into interactive 'vibe-hacking' and the emergence of fully autonomous systems. The paper identifies and discusses significant obstacles impeding wider adoption and safe deployment. These include critical issues concerning model reliability and stability, paramount safety and security concerns, substantial monetary and ecological costs, implications for privacy and digital sovereignty, complex questions of accountability, and profound ethical dilemmas. This comprehensive review and analysis provides a foundation for discussion on future research directions and the development of robust safeguards at the intersection of AI and security.

cs.CR↗

arXiv2025

Benchmarking Practices in LLM-driven Offensive Security: Testbeds, Metrics, and Experiment Design

Large Language Models (LLMs) have emerged as a powerful approach for driving offensive penetration-testing tooling. Due to the opaque nature of LLMs, empirical methods are typically used to analyze their efficacy. The quality of this analysis is highly dependent on the chosen testbed, captured metrics and analysis methods employed. This paper analyzes the methodology and benchmarking practices used for evaluating Large Language Model (LLM)-driven attacks, focusing on offensive uses of LLMs in cybersecurity. We review 19 research papers detailing 18 prototypes and their respective testbeds. We detail our findings and provide actionable recommendations for future research, emphasizing the importance of extending existing testbeds, creating baselines, and including comprehensive metrics and qualitative analysis. We also note the distinction between security research and practice, suggesting that CTF-based challenges may not fully represent real-world penetration testing scenarios.

cs.CR↗

arXiv2025

The Road to Hybrid Quantum Programs: Characterizing the Evolution from Classical to Hybrid Quantum Software

Quantum computing exhibits the unique capability to natively and efficiently encode various natural phenomena, promising theoretical speedups of several orders of magnitude. However, not all computational tasks can be efficiently executed on quantum machines, giving rise to hybrid systems, where some portions of an application run on classical machines, while others utilize quantum resources. Efforts to identify quantum candidate code fragments that can meaningfully execute on quantum machines primarily rely on static code analysis. Yet, the state-of-the-art in static code analysis for quantum candidates remains in its infancy, with limited applicability to specific frameworks and languages, and a lack of generalizability. Existing methods often involve a trial-and-error approach, relying on the intuition and expertise of computer scientists, resulting in varying identification durations ranging from minutes to days for a single application. This paper aims to systematically formalize the process of identifying quantum candidates and their proper encoding within classical programs. Our work addresses the critical initial step in the development of automated reasoning techniques for code-to-code translation, laying the foundation for more efficient quantum software engineering. Particularly, this study investigates a sociotechnical phenomenon where the starting point is not a problem directly solvable with QC, but rather an existing classical program that addresses the problem. In doing so, it underscores the interdisciplinary nature of QC application development, necessitating collaboration between domain experts, computer scientists, and physicists to harness the potential of quantum computing effectively.

cs.SE↗

arXiv2025

Agentic Bug Reproduction for Effective Automated Program Repair at Google

Bug reports often lack sufficient detail for developers to reproduce and fix the underlying defects. Bug Reproduction Tests (BRTs), tests that fail when the bug is present and pass when it has been resolved, are crucial for debugging, but they are rarely included in bug reports, both in open-source and in industrial settings. Thus, automatically generating BRTs from bug reports has the potential to accelerate the debugging process and lower time to repair. This paper investigates automated BRT generation within an industry setting, specifically at Google, focusing on the challenges of a large-scale, proprietary codebase and considering real-world industry bugs extracted from Google's internal issue tracker. We adapt and evaluate a state-of-the-art BRT generation technique, LIBRO, and present our agent-based approach, BRT Agent, which makes use of a fine-tuned Large Language Model (LLM) for code editing. Our BRT Agent significantly outperforms LIBRO, achieving a 28% plausible BRT generation rate, compared to 10% by LIBRO, on 80 human-reported bugs from Google's internal issue tracker. We further investigate the practical value of generated BRTs by integrating them with an Automated Program Repair (APR) system at Google. Our results show that providing BRTs to the APR system results in 30% more bugs with plausible fixes. Additionally, we introduce Ensemble Pass Rate (EPR), a metric which leverages the generated BRTs to select the most promising fixes from all fixes generated by APR system. Our evaluation on EPR for Top-K and threshold-based fix selections demonstrates promising results and trade-offs. For example, EPR correctly selects a plausible fix from a pool of 20 candidates in 70% of cases, based on its top-1 ranking.

cs.SE↗

arXiv2025

Evaluating Agent-based Program Repair at Google

Agent-based program repair offers to automatically resolve complex bugs end-to-end by combining the planning, tool use, and code generation abilities of modern LLMs. Recent work has explored the use of agent-based repair approaches on the popular open-source SWE-Bench, a collection of bugs from highly-rated GitHub Python projects. In addition, various agentic approaches such as SWE-Agent have been proposed to solve bugs in this benchmark. This paper explores the viability of using an agentic approach to address bugs in an enterprise context. To investigate this, we curate an evaluation set of 178 bugs drawn from Google's issue tracking system. This dataset spans both human-reported (78) and machine-reported bugs (100). To establish a repair performance baseline on this benchmark, we implement Passerine, an agent similar in spirit to SWE-Agent that can work within Google's development environment. We show that with 20 trajectory samples and Gemini 1.5 Pro, Passerine can produce a patch that passes bug tests (i.e., plausible) for 73% of machine-reported and 25.6% of human-reported bugs in our evaluation set. After manual examination, we found that 43% of machine-reported bugs and 17.9% of human-reported bugs have at least one patch that is semantically equivalent to the ground-truth patch. These results establish a baseline on an industrially relevant benchmark, which as we show, contains bugs drawn from a different distribution -- in terms of language diversity, size, and spread of changes, etc. -- compared to those in the popular SWE-Bench dataset.

cs.SE↗