Papers for
enterprise cybersecurity teams
Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.
Browser immune system detects web threats using onsite AI model
Browser Immune System (BIS): Innate and Adaptive Threat Detection in the Browser with a Cybersecurity Transformer
Abstract: Security decisions about a web page are best made where the page is: in the browser, before anything leaves the machine. We present the Browser Immune System (BIS), a browser extension that screens navigation with a cybersecurity transformer running entirely on the user's GPU, and alias-web, the model that powers it: a 4B-parameter model that starts from the dual-regime cybersecurity transformer (CT0), extends its public release alias3-micro with a context-category sensor, and is quantised to 4 bits and compiled to a 2.44 GB graph for WebGPU. alias-web is the first in-browser realisation of that model's typed decision mode, System 1, with its official pieces: the marker-token mapping, the complete pointer head and the final normalisation that the exported graph omits. It answers a typed security question in about 150 ms on a laptop GPU, and the same weights generate explanations at 10.5 tokens/s on demand. BIS treats the model as an immune system: an innate response, one System 1 question per threat class, runs on every page; an adaptive response, System 2 text, explains a verdict only when the user asks. On BISBENCH-8, eight synthetic sites of which six execute real but harmless malicious behaviour and two are adversarial controls, BIS returns the expected page verdict on 7/8 sites with both controls green and a class-level micro-F1 of 0.71, scans a page in 0.9 to 6.4 s, and gives identical verdicts across samples and across Chrome and Firefox. Single-variable diagnostics trace both benchmark errors to what the model was shown and, with a first pass over eleven real public pages, show that two of the seven questions answer yes on ordinary scripts and login forms: specificity on the real web is the open problem. We report every error, the engineering obstacles between a research checkpoint and a working extension, and what the evidence does not yet support.
Agentic AI systems gain full protection from hidden prompt attacks
Agentic AI Defense: Decode-Time Provenance Binding for MCP
Abstract: Agentic AI systems using the Model Context Protocol (MCP) are vulnerable to indirect prompt injection through untrusted tool outputs. We demonstrate cross-modal trust laundering, in which instructions concealed in an email-signature image using least-significant-bit (LSB), discrete cosine transform (DCT), or metadata steganography bypass text scanners and enter a multi-agent workflow when decoded by a tool. To counter this threat, we propose decode-time provenance binding, which combines HMAC-authenticated trust labels with an MCP-maintained provenance ledger and context-taint tracking. The mechanism labels decoded content as untrusted and blocks critical tool calls influenced by that content, even when a compromised agent substitutes a valid trusted label. We evaluate the approach using a real MCP testbed with three steganographic carriers and 24 adaptive payloads. The undefended pipeline exhibits 100% attack success, while keyword and regex defenses each permit 62.5% of attacks. Our mechanism blocks all attacks (0% success), including adversarial label substitution across mediated agent boundaries. A four-component ablation validates the roles of decode marking, cryptographic authentication, lineage tracking, and sink enforcement. An audited trusted-promotion mechanism increases legitimate-task completion from 67% to 100% without compromising protection. The approach provides carrier-independent containment within fully mediated MCP execution environments. Code and data are publicly available.
Llm email agents fail to detect and resist adaptive phishing attacks
AgentPhishing: A Benchmark of LLM Email Agents Under Adaptive Attack
Abstract: Computer-use agents increasingly handle private information and act on users' behalf across desktop and web applications, including email. Phishing therefore poses privacy and security risks that depend not only on whether agents recognize malicious messages, but also on how they respond. Existing phishing datasets were developed primarily for conventional detection, leaving agent susceptibility to persuasive malicious requests insufficiently characterized. To expose vulnerabilities that pre-existing datasets may miss, we introduce AgentPhishing, a benchmark for evaluating whether persuasive phishing attacks can evade detection and induce LLM email agents to comply with malicious requests. We develop PhishOpt, which optimizes reusable email-generation skills against individual agents under fixed benchmark constraints and a realism constraint. We analyze the generated emails using Cialdini's persuasion principles to identify strategies associated with agent failures and use these findings to construct a universal generation skill. Our evaluation shows that agents can handle pre-existing phishing emails safely yet remain susceptible to attacks adapted to agent behavior, while in an AgentPhishing evaluation of nine agents, four exhibit phishing-detection rates below $10\%$ and malicious-request compliance rates above $45\%$. These results highlight the limits of inferring agent safety from conventional phishing evaluations and the need to assess how agents respond to persuasive malicious requests.
Adaptive cyber defense improves network security against evolving attacks
Adaptive Cyber-Defense for Evolving Attacker Capabilities using Reinforcement Learning and Behavior Trees
Abstract: Autonomous cyber-defense agents trained using reinforcement learning (RL) are increasingly adopted for defending enterprise computer networks. However, existing approaches may become brittle as attacker capabilities evolve and enable new attack strategies. In this work, we study the problem of adapting defender policies learned in CAGE Challenge Scenario 2 (CC2), a benchmark enterprise network defense scenario, to an extended network scenario with new attacker capabilities represented by new attacker actions. Effective attack strategies that leverage new actions are initially unknown, and therefore attacker policies must be learned before defender adaptation can occur. To address this challenge, we develop adversarial RL (ARL) with behavior trees (ARLBT), which uses reward behavior trees (RBTs) to define progression-based reward functions and condition attacker policies on RBT execution. The proposed RBT incorporates the new attacker actions into structured attack strategies while leveraging parallel execution to generate rewards across multiple network nodes simultaneously. We evaluate ARLBT for learning attacker policies against two baseline defender policies across network scenarios with different sets of new attacker actions and compare against multiple attacker baselines. The results demonstrate that ARLBT improves attacker success compared to ARL using the original CC2 reward function and that conditioning policies on RBT execution can improve sample efficiency. We further show that defender policies can be adapted against the learned attackers, improving robustness to attacks that leverage new attacker actions.
Multi-vector visual document indices can leak readable pages
Inverting Multi-Vector Visual Document Indices
Abstract: Prevailing multi-vector visual document retrievers store each page as about a thousand patch vectors, often in vector databases run by a third party. Since no one can read a page from its vectors, this index is easily treated as less sensitive than the page. However, because the index keeps one vector per patch in raster order, and each vector is computed by a vision-language model pre-trained to read documents, we hypothesize that whoever runs or breaches the store can reproduce a page from its index alone. We frame inversion as conditional document image generation and infer from the vectors what the attack needs: the encoder, the page shape and, for shuffled vectors, their order. On the ViDoRe v3 benchmark, pages inverted from raw indices recover 47% of the words and 45% of the sensitive tokens. Used as queries against the stored indices, they rank their source page first 98.4% of the time. We test two cheap protections, token pooling and shuffling, which both cut word recall to about 8%. A model that restores the order of a shuffled index raises the share of source pages ranked first from 3.8% to 93.5%, while inverting a pooled index remains open. To test generalisation, we apply the same attack unchanged to another multi-vector retriever: its inverted pages still rank their source page first 70.2% of the time, though its word recall stays below a nearest-neighbour baseline. Multi-vector visual document retrievers are therefore vulnerable to inversion through their stored index, which should be protected like the documents it encodes.
Ai helps prioritize cyberattack risks in industrial networks
AI-Based Vulnerability Assessment Capability and Cyber Attack Graph Analysis
Abstract: Cyber threats targeting mission-critical infrastructure are becoming more sophisticated while the barrier to launching attacks continues to fall. Traditional point solutions like antivirus and firewalls are reactive and fail to address the combinatorial complexity of modern attack surfaces. This paper presents an investigation combining two complementary methodologies: Lockheed Martin's Vortex/Crow framework, which applies multi-agent reinforcement learning (MARL) over industry-standard cyber knowledge graph to identify and prioritize attack vectors and TTPs (tactics, techniques, and procedures); and Aalto's probabilistic attack graph model that combines network topology and its vulnerabilities to compute system-level risk metrics. The 2015 Ukraine Power Grid cyberattack serves as a well-documented validation scenario. Applied independently to the same operational technology (OT) network topology, both methodologies converge on the same attack vectors and exploit sequences as those documented in the incident record, thus providing mutual cross-validation. Attack graph analyses using node-level elimination experiments identify industrial control systems (ICS) as the most critical enablers of attack propagation, representing high-priority targets for defensive hardening. Comparison of CVSS (v2.0) and IronMiner vulnerability scoring yields in general consistent results, with IronMiner providing more actionable differentiation at network periphery nodes. The layered methodology of baseline assessment and node-level elimination proves to be scalable to large enterprise networks, thus offering defenders a structured, AI-enabled path to prioritize mitigation under realistic time and resource constraints.
Poisoning malware detectors by exploiting antivirus label weaknesses
Weaponizing Ground Truth: Data Poisoning Attacks by Exploiting Boundary Misalignment Between Antivirus Software and Learning-Based Detectors
Abstract: Machine-learning (ML)-based malware detectors are commonly trained using labels obtained from antivirus (AV) engines and aggregation services (e.g., VirusTotal). This practice assumes AV-generated labels provide reliable supervision. However, small byte-level modifications can substantially alter AV verdicts while leaving the representations perceived by downstream ML detectors largely unchanged, producing label-feature inconsistencies that can contaminate training datasets and create poisoning opportunities for ML-based malware detection. We present Bi-Iocane, a black-box poisoning framework that exploits the reliance of malware-labeling pipelines on AV-generated labels. Bi-Iocane identifies AV-sensitive bytes and modifies them to induce label changes. It rewrites such bytes in malware to obtain benign labels (evasion-oriented poisoning) and injects malware-associated byte patterns into benign software to obtain malicious labels (defamation-oriented poisoning). These poisoned samples and their lightly modified variants corrupt training data and cause selected targets to be misclassified. We evaluate Bi-Iocane with 13 AV engines simulating AV aggregation services and eight ML detectors. For 30 malware and 30 benign clean targets, Bi-Iocane combines AV-specific manipulations to generate malware-to-benign and benign-to-malware poisoned samples whose all tested AV-based labels are flipped. After these poisoned samples and variants are used for downstream training, the resulting ML models misclassify 92.08% of the original clean targets on average with only a 0.06\% poisoning budget per target. Meanwhile, the poisoned models largely preserve clean-set performance, and six evaluated poisoning defenses show only limited mitigation. VirusTotal evaluation further confirms practical defamation risk and reveals potential evasion risk in real-world AV-to-ML labeling supply chains.