arXiv subjects

Andrew Paverd

Publications and source records attributed to Andrew Paverd.

At least 19 recordsLinked to original sources

arXiv2026

SpecGuard: Inference-Time Backdoor Detection For Free

Large language models are often fine-tuned, shared, or downloaded from third parties, so a deployed model may carry a hidden backdoor that behaves normally on benign inputs but switches to attacker-controlled behavior when a secret trigger appears. While backdoors can be audited before deployment, runtime monitoring remains important for models that are frequently updated. The challenge is that LLM serving is latency-sensitive: existing inference-time detectors either rely on assumptions about the trigger form, which can fail on stealthy attacks, or require extra model computation, such as input perturbations or an additional generation pass. We introduce SpecGuard, an inference-time backdoor detector that repurposes speculative decoding at zero added model-computation cost. Speculative decoding speeds up inference by using a small draft model to propose tokens and a target model to verify them. We observe that this verification process already exposes a useful signal: when a backdoor is triggered, the target model shifts toward the attacker's behavior, while a clean draft model does not predict this shift, causing the draft-token acceptance rate to change. We formalize when this signal appears and show that an attacker who suppresses it must also weaken the backdoor. Across diverse backdoor types and model families, SpecGuard reliably detects triggered behavior, including stealthy cases where input-level filters are blind, while avoiding the extra generation cost of existing runtime detectors. Speculative decoding therefore doubles as a free, always-on signal for detecting backdoored LLM behavior.

cs.CR↗

arXiv2026

Rising From the Ashes: How Agentic AI is Unblocking Challenges in Cybersecurity

Security remains a high-cost challenge, with many problems historically deemed inefficient to address or effectively unsolvable. A significant number of these problems stem from labor-intensive tasks that create bottlenecks in defensive approaches. Agentic AI has the potential to alleviate these bottlenecks by directly ingesting and reasoning over natural language or code, thereby expanding the scope of feasible defenses. In this paper, we map open security problems to emergent agentic AI capabilities. To illustrate this potential, we examine 16 case studies, including supply chain analysis, highlighting how agentic AI may benefit defenders.

cs.CR↗

arXiv2026

MetaBackdoor: Exploiting Positional Encoding as a Backdoor Attack Surface in LLMs

Backdoor attacks pose a serious security threat to large language models (LLMs), which are increasingly deployed as general-purpose assistants in safety- and privacy-critical applications. Existing LLM backdoors rely primarily on content-based triggers, requiring explicit modification of the input text. In this work, we show that this assumption is unnecessary and limiting. We introduce MetaBackdoor, a new class of backdoor attacks that exploits positional information as the trigger, without modifying textual content. Our key insight is that Transformer-based LLMs necessarily encode token positions to process ordered sequences. As a result, length-correlated positional structure is reflected in the model's internal computation and can be used as an effective non-content trigger signal. We demonstrate that even a simple length-based positional trigger is sufficient to activate stealthy backdoors. Unlike prior attacks, MetaBackdoor operates on visibly and semantically clean inputs and enables qualitatively new capabilities. We show that a backdoored LLM can be induced to disclose sensitive internal information, including proprietary system prompts, once a length condition is satisfied. We further demonstrate a self-activation scenario, where normal multi-turn interaction can move the conversation context into the trigger region and induce malicious tool-call behavior without attacker-supplied trigger text. In addition, MetaBackdoor is orthogonal to content-based backdoors and can be composed with them to create more precise and harder-to-detect activation conditions. Our results expand the threat model of LLM backdoors by revealing positional encoding as a previously overlooked attack surface. This challenges defenses that focus on detecting suspicious text and highlights the need for new defense strategies that explicitly account for positional triggers in modern LLM architectures.

cs.CR↗

arXiv2026

Stateless Yet Not Forgetful: Implicit Memory as a Hidden Channel in LLMs

Large language models (LLMs) are commonly treated as stateless: once an interaction ends, no information is assumed to persist unless it is explicitly stored and re-supplied. We challenge this assumption by introducing implicit memory-the ability of a model to carry state across otherwise independent interactions by encoding information in its own outputs and later recovering it when those outputs are reintroduced as input. This mechanism does not require any explicit memory module, yet it creates a persistent information channel across inference requests. As a concrete demonstration, we introduce a new class of temporal backdoors, which we call time bombs. Unlike conventional backdoors that activate on a single trigger input, time bombs activate only after a sequence of interactions satisfies hidden conditions accumulated via implicit memory. We show that such behavior can be induced today through straightforward prompting or fine-tuning. Beyond this case study, we analyze broader implications of implicit memory, including covert inter-agent communication, benchmark contamination, targeted manipulation, and training-data poisoning. Finally, we discuss detection challenges and outline directions for stress-testing and evaluation, with the goal of anticipating and controlling future developments. To promote future research, we release code and data at: https://github.com/microsoft/implicitMemory.

cs.LG↗

arXiv2025

Highlight & Summarize: RAG without the jailbreaks

Preventing jailbreaking and model hijacking of Large Language Models (LLMs) is an important yet challenging task. When interacting with a chatbot, malicious users can input specially crafted prompts that cause the LLM to generate undesirable content or perform a different task from its intended purpose. Existing systems attempt to mitigate this by hardening the LLM's system prompt or using additional classifiers to detect undesirable content or off-topic conversations. However, these probabilistic approaches are relatively easy to bypass due to the very large space of possible inputs and undesirable outputs. We present and evaluate Highlight & Summarize (H&S), a new design pattern for retrieval-augmented generation (RAG) systems that prevents these attacks by design. The core idea is to perform the same task as a standard RAG pipeline (i.e., to provide natural language answers to questions, based on relevant sources) without ever revealing the user's question to the generative LLM. This is achieved by splitting the pipeline into two components: a highlighter, which takes the user's question and extracts ("highlights") relevant passages from the retrieved documents, and a summarizer, which takes the highlighted passages and summarizes them into a cohesive answer. We describe and implement several possible instantiations of H&S and evaluate their responses in terms of correctness, relevance, and quality. For certain question-answering (QA) tasks, the responses produced by H&S are judged to be as good, if not better, than those of a standard RAG pipeline.

cs.CL↗

arXiv2025

LLMail-Inject: A Dataset from a Realistic Adaptive Prompt Injection Challenge

Indirect Prompt Injection attacks exploit the inherent limitation of Large Language Models (LLMs) to distinguish between instructions and data in their inputs. Despite numerous defense proposals, the systematic evaluation against adaptive adversaries remains limited, even when successful attacks can have wide security and privacy implications, and many real-world LLM-based applications remain vulnerable. We present the results of LLMail-Inject, a public challenge simulating a realistic scenario in which participants adaptively attempted to inject malicious instructions into emails in order to trigger unauthorized tool calls in an LLM-based email assistant. The challenge spanned multiple defense strategies, LLM architectures, and retrieval configurations, resulting in a dataset of 208,095 unique attack submissions from 839 participants. We release the challenge code, the full dataset of submissions, and our analysis demonstrating how this data can provide new insights into the instruction-data separation problem. We hope this will serve as a foundation for future research towards practical structural solutions to prompt injection.

cs.CR↗

arXiv2025

Design Patterns for Securing LLM Agents against Prompt Injections

As AI agents powered by Large Language Models (LLMs) become increasingly versatile and capable of addressing a broad spectrum of tasks, ensuring their security has become a critical challenge. Among the most pressing threats are prompt injection attacks, which exploit the agent's resilience on natural language inputs -- an especially dangerous threat when agents are granted tool access or handle sensitive information. In this work, we propose a set of principled design patterns for building AI agents with provable resistance to prompt injection. We systematically analyze these patterns, discuss their trade-offs in terms of utility and security, and illustrate their real-world applicability through a series of case studies.

cs.LG↗

arXiv2025

Securing AI Agents with Information-Flow Control

As AI agents become increasingly autonomous and capable, ensuring their security against vulnerabilities such as prompt injection becomes critical. This paper explores the use of information-flow control (IFC) to provide security guarantees for AI agents. We present a formal model to reason about the security and expressiveness of agent planners. Using this model, we characterize the class of properties enforceable by dynamic taint-tracking and construct a taxonomy of tasks to evaluate security and utility trade-offs of planner designs. Informed by this exploration, we present Fides, a planner that tracks confidentiality and integrity labels, deterministically enforces security policies, and introduces novel primitives for selectively hiding information. Its evaluation in AgentDojo demonstrates that this approach enables us to complete a broad range of tasks with security guarantees. A tutorial to walk readers through the the concepts introduced in the paper can be found at https://github.com/microsoft/fides

cs.CR↗

arXiv2024

Practice-Informed, Practice-Ready: An AI security incident taxonomy

With the increasing prevalence of AI systems, several real-world AI security incidents have been reported. However, despite forthcoming legal mandates, the reporting and collection of these incidents still lacks practical standards and proposals. We bridge this gap by establishing a rigorous foundation based on discussions with a diverse group of AI practitioners spanning industrial, non-profit, research, and governmental sectors. Our proposed taxonomy provides concrete guidance to identify affected parties, recommend relevant security measures, and gain an actionable overview of the evolving AI security landscape. Our tests show that different coders consistently identify similar topics, but that automating incident tagging via an LLM like ChatGPT is of limited use. Notably, our framework has already served as the scientific basis for an established industry standard, proving its utility and readiness for widespread adoption.

cs.CR↗

arXiv2024

VerifiableFL: Verifiable Claims for Federated Learning using Exclaves

In federated learning (FL), data providers jointly train a machine learning model without sharing their training data. This makes it challenging to provide verifiable claims about the trained FL model, e.g., related to the employed training data, any data sanitization, or the correct training algorithm-a malicious data provider can simply deviate from the correct training protocol without detection. While prior FL training systems have explored the use of trusted execution environments (TEEs) to protect the training computation, such approaches rely on the confidentiality and integrity of TEEs. The confidentiality guarantees of TEEs, however, have been shown to be vulnerable to a wide range of attacks, such as side-channel attacks. We describe VerifiableFL, a system for training FL models that establishes verifiable claims about trained FL models with the help of fine-grained runtime attestation proofs. Since these runtime attestation proofs only require integrity protection, VerifiableFL generates them using the new abstraction of exclaves. Exclaves are integrity-only execution environments, which do not contain software-managed secrets and thus are immune to data leakage attacks. VerifiableFL uses exclaves to attest individual data transformations during FL training without relying on confidentiality guarantees. The runtime attestation proofs then form an attested dataflow graph of the entire FL model training computation. The graph is checked by an auditor to ensure that the trained FL model satisfies its claims, such as the use of data sanitization by data providers or correct aggregation by the model provider. VerifiableFL extends NVFlare FL framework to use exclaves. We show that VerifiableFL introduces less than 12% overhead compared to unprotected FL training.

cs.CR↗

arXiv2024

Permissive Information-Flow Analysis for Large Language Models

Large Language Models (LLMs) are rapidly becoming commodity components of larger software systems. This poses natural security and privacy problems: poisoned data retrieved from one component can change the model's behavior and compromise the entire system, including coercing the model to spread confidential data to untrusted components. One promising approach is to tackle this problem at the system level via dynamic information flow (aka taint) tracking. Unfortunately, this approach of propagating the most restrictive input label to the output is too conservative for applications where LLMs operate on inputs retrieved from diverse sources. In this paper, we propose a novel, more permissive approach to propagate information flow labels through LLM queries. The key idea behind our approach is to propagate only the labels of the samples that were influential in generating the model output and to eliminate the labels of unnecessary inputs. We implement and investigate the effectiveness of two variations of this approach, based on (i) prompt-based retrieval augmentation, and (ii) a $k$-nearest-neighbors language model. We compare these with a baseline that uses introspection to predict the output label. Our experimental results in an LLM agent setting show that the permissive label propagator improves over the baseline in more than 85% of the cases, which underscores the practicality of our approach.

cs.LG↗

arXiv2024

Get my drift? Catching LLM Task Drift with Activation Deltas

LLMs are commonly used in retrieval-augmented applications to execute user instructions based on data from external sources. For example, modern search engines use LLMs to answer queries based on relevant search results; email plugins summarize emails by processing their content through an LLM. However, the potentially untrusted provenance of these data sources can lead to prompt injection attacks, where the LLM is manipulated by natural language instructions embedded in the external data, causing it to deviate from the user's original instruction(s). We define this deviation as task drift. Task drift is a significant concern as it allows attackers to exfiltrate data or influence the LLM's output for other users. We study LLM activations as a solution to detect task drift, showing that activation deltas - the difference in activations before and after processing external data - are strongly correlated with this phenomenon. Through two probing methods, we demonstrate that a simple linear classifier can detect drift with near-perfect ROC AUC on an out-of-distribution test set. We evaluate these methods by making minimal assumptions about how users' tasks, system prompts, and attacks can be phrased. We observe that this approach generalizes surprisingly well to unseen task domains, such as prompt injections, jailbreaks, and malicious instructions, without being trained on any of these attacks. Interestingly, the fact that this solution does not require any modifications to the LLM (e.g., fine-tuning), as well as its compatibility with existing meta-prompting solutions, makes it cost-efficient and easy to deploy. To encourage further research on activation-based task inspection, decoding, and interpretability, we release our large-scale TaskTracker toolkit, featuring a dataset of over 500K instances, representations from six SoTA language models, and a suite of inspection tools.

cs.CR↗

arXiv2024

Closed-Form Bounds for DP-SGD against Record-level Inference

Machine learning models trained with differentially-private (DP) algorithms such as DP-SGD enjoy resilience against a wide range of privacy attacks. Although it is possible to derive bounds for some attacks based solely on an $(\varepsilon,\delta)$-DP guarantee, meaningful bounds require a small enough privacy budget (i.e., injecting a large amount of noise), which results in a large loss in utility. This paper presents a new approach to evaluate the privacy of machine learning models against specific record-level threats, such as membership and attribute inference, without the indirection through DP. We focus on the popular DP-SGD algorithm, and derive simple closed-form bounds. Our proofs model DP-SGD as an information theoretic channel whose inputs are the secrets that an attacker wants to infer (e.g., membership of a data record) and whose outputs are the intermediate model parameters produced by iterative optimization. We obtain bounds for membership inference that match state-of-the-art techniques, whilst being orders of magnitude faster to compute. Additionally, we present a novel data-dependent bound against attribute inference. Our results provide a direct, interpretable, and practical way to evaluate the privacy of trained models against specific inference threats without sacrificing utility.

cs.CR↗

arXiv2023

Maatphor: Automated Variant Analysis for Prompt Injection Attacks

Prompt injection has emerged as a serious security threat to large language models (LLMs). At present, the current best-practice for defending against newly-discovered prompt injection techniques is to add additional guardrails to the system (e.g., by updating the system prompt or using classifiers on the input and/or output of the model.) However, in the same way that variants of a piece of malware are created to evade anti-virus software, variants of a prompt injection can be created to evade the LLM's guardrails. Ideally, when a new prompt injection technique is discovered, candidate defenses should be tested not only against the successful prompt injection, but also against possible variants. In this work, we present, a tool to assist defenders in performing automated variant analysis of known prompt injection attacks. This involves solving two main challenges: (1) automatically generating variants of a given prompt according, and (2) automatically determining whether a variant was effective based only on the output of the model. This tool can also assist in generating datasets for jailbreak and prompt injection attacks, thus overcoming the scarcity of data in this domain. We evaluate Maatphor on three different types of prompt injection tasks. Starting from an ineffective (0%) seed prompt, Maatphor consistently generates variants that are at least 60% effective within the first 40 iterations.

cs.CR↗

arXiv2023

Rethinking Privacy in Machine Learning Pipelines from an Information Flow Control Perspective

Modern machine learning systems use models trained on ever-growing corpora. Typically, metadata such as ownership, access control, or licensing information is ignored during training. Instead, to mitigate privacy risks, we rely on generic techniques such as dataset sanitization and differentially private model training, with inherent privacy/utility trade-offs that hurt model performance. Moreover, these techniques have limitations in scenarios where sensitive information is shared across multiple participants and fine-grained access control is required. By ignoring metadata, we therefore miss an opportunity to better address security, privacy, and confidentiality challenges. In this paper, we take an information flow control perspective to describe machine learning systems, which allows us to leverage metadata such as access control policies and define clear-cut privacy and confidentiality guarantees with interpretable information flows. Under this perspective, we contrast two different approaches to achieve user-level non-interference: 1) fine-tuning per-user models, and 2) retrieval augmented models that access user-specific datasets at inference time. We compare these two approaches to a trivially non-interfering zero-shot baseline using a public model and to a baseline that fine-tunes this model on the whole corpus. We evaluate trained models on two datasets of scientific articles and demonstrate that retrieval augmented architectures deliver the best utility, scalability, and flexibility while satisfying strict non-interference guarantees.

cs.LG↗

arXiv2023

An Empirical Study & Evaluation of Modern CAPTCHAs

For nearly two decades, CAPTCHAs have been widely used as a means of protection against bots. Throughout the years, as their use grew, techniques to defeat or bypass CAPTCHAs have continued to improve. Meanwhile, CAPTCHAs have also evolved in terms of sophistication and diversity, becoming increasingly difficult to solve for both bots (machines) and humans. Given this long-standing and still-ongoing arms race, it is critical to investigate how long it takes legitimate users to solve modern CAPTCHAs, and how they are perceived by those users. In this work, we explore CAPTCHAs in the wild by evaluating users' solving performance and perceptions of unmodified currently-deployed CAPTCHAs. We obtain this data through manual inspection of popular websites and user studies in which 1,400 participants collectively solved 14,000 CAPTCHAs. Results show significant differences between the most popular types of CAPTCHAs: surprisingly, solving time and user perception are not always correlated. We performed a comparative study to investigate the effect of experimental context -- specifically the difference between solving CAPTCHAs directly versus solving them as part of a more natural task, such as account creation. Whilst there were several potential confounding factors, our results show that experimental context could have an impact on this task, and must be taken into account in future CAPTCHA studies. Finally, we investigate CAPTCHA-induced user task abandonment by analyzing participants who start and do not complete the task.

cs.CR↗

arXiv2023

On the Efficacy of Differentially Private Few-shot Image Classification

There has been significant recent progress in training differentially private (DP) models which achieve accuracy that approaches the best non-private models. These DP models are typically pretrained on large public datasets and then fine-tuned on private downstream datasets that are relatively large and similar in distribution to the pretraining data. However, in many applications including personalization and federated learning, it is crucial to perform well (i) in the few-shot setting, as obtaining large amounts of labeled data may be problematic; and (ii) on datasets from a wide variety of domains for use in various specialist settings. To understand under which conditions few-shot DP can be effective, we perform an exhaustive set of experiments that reveals how the accuracy and vulnerability to attack of few-shot DP image classification models are affected as the number of shots per class, privacy level, model architecture, downstream dataset, and subset of learnable parameters in the model vary. We show that to achieve DP accuracy on par with non-private models, the shots per class must be increased as the privacy level increases. We also show that learning parameter-efficient FiLM adapters under DP is competitive with learning just the final classifier layer or learning all of the network parameters. Finally, we evaluate DP federated learning systems and establish state-of-the-art performance on the challenging FLAIR benchmark.

stat.ML↗

arXiv2022

SoK: Let the Privacy Games Begin! A Unified Treatment of Data Inference Privacy in Machine Learning

Deploying machine learning models in production may allow adversaries to infer sensitive information about training data. There is a vast literature analyzing different types of inference risks, ranging from membership inference to reconstruction attacks. Inspired by the success of games (i.e., probabilistic experiments) to study security properties in cryptography, some authors describe privacy inference risks in machine learning using a similar game-based style. However, adversary capabilities and goals are often stated in subtly different ways from one presentation to the other, which makes it hard to relate and compose results. In this paper, we present a game-based framework to systematize the body of knowledge on privacy inference risks in machine learning. We use this framework to (1) provide a unifying structure for definitions of inference risks, (2) formally establish known relations among definitions, and (3) to uncover hitherto unknown relations that would have been difficult to spot otherwise.

cs.LG↗