Memory systems enable otherwise-stateless LLM agents to persist user information across sessions, but also introduce a new attack surface. We characterize the Trojan Hippo attack, a class of persistent memory attacks that operates in a more realistic threat model than prior memory poisoning work: the attacker plants a dormant payload into an agent's long-term memory via a single untrusted tool call (e.g., a crafted email), which activates only when the user later discusses sensitive topics such as finance, health, or identity, and exfiltrates high-value personal data to the attacker. While anecdotal demonstrations of such attacks have appeared against deployed systems, no prior work systematically evaluates them across heterogeneous memory architectures and defenses.We introduce a dynamic evaluation framework comprising two components: (1) an OpenEvolve-based adaptive red-teaming benchmark that stress-tests defenses and memory backends against continuously refined attacks, and (2) the first capability-aware security/utility analysis for persistent memory systems, enabling principled reasoning about defense deployment across different usage profiles. Instantiated on an email assistant across four memory backends (explicit tool memory, agentic memory, RAG, and sliding-window context), Trojan Hippo achieves up to 85-100 percent ASR against current frontier models from OpenAI and Google, with planted memories successfully activating even after 100 benign sessions. We evaluate four memory-system defenses inspired by basic security principles, finding they substantially reduce attack success rates (to as low as 0-5 percent), though at utility costs that vary widely with task requirements. Because of this substantial security-utility tradeoff, the effective real-world deployment of defenses remains an open challenge, which our evaluation framework is specifically designed to address.

Prompt injection is widely recognized as a major security threat to AI agents that interact with untrusted external data, such as websites, documents, and emails. Prior work has shown that, in the text domain, black-box prompt injection can achieve near-perfect attack success rates (ASRs). In the image domain, however, existing visual prompt injection methods are substantially less effective in attacking frontier commercial VLMs for materially harmful behavior. Achieving such outputs is hard because it requires a long and/or format-compliant target string, such as a precise, parseable native tool call with exact function names and arguments. We present Repeat-After-Me, a black-box adaptive visual prompt injection attack that can reveal personally identifiable information or make malicious tool calls. Across both open-weight and commercial frontier VLMs, including Qwen3.6-27B and GPT-5.5, our method achieves ASRs exceeding 80

Security analysts routinely query system logs to detect threats and investigate incidents, but each log source uses its own semi-structured format: logs are cheap to produce, but expensive to use. The standard approach, building per-source parsers to normalize logs into structured schemas, is powerful but requires continuous engineering effort for each new format. Querying raw logs directly with tools like grep avoids this cost, but requires analysts to know each source's message variants and cannot express the multi-line temporal queries that security investigations demand. We present Sieve, a system that generates executable query code from natural-language security questions by grounding a large language model with lightweight, automatically extracted log-format context, requiring only one LLM call per query followed by deterministic execution. Evaluating 133 security queries across 5 log types, we find that Sieve achieves over a 3x reduction in error rate on complex temporal and cross-event queries compared to manual analyst scripting, with the largest gains on the multi-line correlation tasks most critical to active investigations. Our results and benchmark provide evidence that LLM-generated code can bridge the gap between the expressiveness of structured log querying and the immediacy of working directly with raw files.

ReAct has become the default architecture across LLM agents, and many existing web agents follow this paradigm. We argue that it is the wrong default for web agents. Instead, web agents should default to plan-then-execute: commit to a task-specific program before observing runtime web content, then execute it. The reason is that web content mixes inputs from many parties. An e-commerce product page may combine a seller's listing, customer reviews and sponsored advertisements. Under ReAct, all of this content flows into the model when deciding on the next action, creating a direct path for prompt injections to steer the agent's control flow. Plan-then-execute changes this boundary: untrusted data may influence values or branches inside a predefined execution graph, but it cannot redefine the user task or cause the model to synthesize new actions at runtime. We analyze WebArena, a popular web agent benchmark, and find that all tasks are compatible with plan-then-execute, while 80

Memory safety vulnerabilities remain a significant threat even for projects with extensive fuzzing and manual auditing. Recent results suggest that large language models hold great promise for detecting such vulnerabilities, but they are unreliable, at risk of hallucination, and challenging to scale to repository-size codebases. This paper presents Revelio, a cost-efficient end-to-end agentic framework for memory-safety vulnerability discovery. Revelio addresses the problem of hallucination by generating an executable Proof-of-Vulnerability, which is checked with a deterministic sanitizer. It reduces cost using inexpensive LLMs and lightweight static analysis to help generate and rank vulnerability hypotheses, reporting vulnerabilities only when they can be reproduced and confirmed by a sanitizer. We evaluated Revelio on seven production-quality projects that had been continuously fuzzed for five to eight years, as well as on 100 randomly selected Arvo projects from the CyberGym benchmark. With around one hour per project and a total cost of $300, Revelio discovered 19 previously unknown memory-safety vulnerabilities. On benchmarks, Revelio outperformed frontier coding agents across diverse backbone models at comparable token costs. Our results suggest that Revelio enables scalable and trustworthy end-to-end LLM-based memory-safety vulnerability detection.

Prompt injection is listed as the #1 threat to AI agents. When an agent accesses external data from websites, files, or emails, an attacker may inject a prompt into the data, saying, "Ignore all prior instructions and perform ." To prevent arbitrary manipulation of agents, defenders try to train secure LLMs, which, however, still suffer from near 100

Large Language Models (LLMs) pose a significant risk of safety misalignment after finetuning, as models can be compromised by both explicitly and implicitly harmful data. Even some seemingly benign data can inadvertently steer a model towards misaligned behaviors. To address this, we introduce GradShield, a principled filtering method that safeguards LLMs during finetuning by identifying and removing harmful data points before they corrupt the model's alignment. It removes potentially harmful data by computing a Finetuning Implicit Harmfulness Score (FIHS) for each data point and employs an adaptive thresholding algorithm. We apply GradShield to multiple utility fine-tuning tasks across varying levels of harmful data and evaluate the safety and utility performance of the resulting LLMs using various metrics. The results show that GradShield outperforms all baseline methods, consistently maintaining an Attack Success Rate (ASR) below 6% while preserving utility performance.

Large language model (LLM) agents are vulnerable to security risks, such as prompt injection attacks from untrusted context that manipulate downstream reasoning and tool use. Existing secure-by-design approaches mitigate this risk by separating untrusted observations from privileged execution and careful control of information flow, but often degrade utility and require extensive task-specific engineering. We thus propose Twin Agent, a general privilege separation design pattern inspired by residual coding in the agent context. Twin Agent consists of two nearly symmetric agents: an Explore Agent that inspects untrusted information and a Safe Agent that executes privileged actions. The Explore Agent is conditioned on the Safe Agent's current context and communicates only compact hints to the Safe Agent about the next action to take. This design reduces the information needed to preserve task utility and thus achieves a better security–utility tradeoff, which we empirically verify by measuring how utility and attack success change as the length of hints varies. We evaluate Twin Agent on long-horizon software engineering tasks with SWE-bench Lite and on heterogeneous multi-tool interaction tasks with AgentDojo and DecodingTrust-Agent. Across both benchmarks, Twin Agent preserves high task utility while preventing prompt injection attacks, outperforming both undefended agents and privilege separation baselines.

Large Language Models (LLMs) are trained with safety alignment to prevent generating malicious content. Although some attacks have highlighted vulnerabilities in these safety-aligned LLMs, they typically have limitations, such as necessitating access to the model weights or the generation process. Since proprietary models through API-calling do not grant users such permissions, these attacks find it challenging to compromise them. In this paper, we propose Jailbreaking Using LLM Introspection (JULI), which jailbreaks LLMs by manipulating the token log probabilities, using a tiny plug-in block, BiasNet. JULI relies solely on the knowledge of the target LLM's predicted token log probabilities. It can effectively jailbreak API-calling LLMs under a black-box setting and knowing only top-5 token log probabilities. Our approach demonstrates superior effectiveness, outperforming existing state-of-the-art (SOTA) approaches across multiple metrics.

Traditionally, machine learning methods for PE malware detection have relied on static features like byte histograms, string information, and PE header contents. One barrier to incorporating dynamic analysis features has been the semi-structured nature of sandbox behavior reports. We show that, using the latest generation of large language models with reasoning, it is possible to efficiently process these behavior reports and utilize them as part of a malware detection pipeline. Specifically, we leverage LLMs to generate behavior-based malware detection rules based on a small training set of labeled malware. We find that these detection rules, derived from behavioral features, are much more robust to concept drift than standard static-feature methods, while maintaining practical false positive rates. Finally, we introduce Trident, a system which combines a classic decision tree model over static features, our behavior-based detection rules, and direct LLM analysis of sandbox reports through majority voting. Trident outperforms standard methods using static features, outperforms behavior-based rules alone, and is as resilient to concept drift as active learning methods without requiring retraining.

Cross-Site Scripting (XSS) remains one of the most prevalent and damaging classes of web vulnerabilities. LLM-based coding agents offer a promising approach to XSS discovery by combining source-code reasoning with interactive testing against a running application. However, a coding agent's claims cannot be trusted on their own. We characterize three reward-hacking behaviors in white-box agentic XSS discovery and propose three requirements that an ideal verifier should meet. We present RECEIPT, a verification framework that makes agent-reported XSS findings trustworthy by enforcing environment isolation, PoC constraints, role separation, and verdict binding. Each confirmation therefore establishes two properties: the script runs in a real browser, and the payload was planted under the attacker role and executed in the victim role's browser. This constrained replay procedure makes validation deterministic and reproducible. We evaluate RECEIPT on 95 real-world web-application targets drawn from popular open-source projects. Within a $20 per-application budget, RECEIPT found 24 previously unknown XSS vulnerabilities, 12 of which have already been acknowledged by maintainers after responsible disclosure, and recovered the labeled CVE in 36% of known-vulnerability recovery targets. Compared with the same agent using self-judgment and with black-box scanners, RECEIPT confirms more real exploits while admitting no false positives.

Prompt injection attacks pose a significant security threat to LLM-integrated applications. Model-level defenses have shown strong effectiveness, but are currently deployed into commercial-grade models in a closed-source manner. We believe open-source models are needed by the AI security community, where co-development of attacks and defenses through open research drives scientific progress in mitigation against prompt injection attacks. To this end, we develop Meta SecAlign, the first open-source and open-weight LLM with built-in model-level defense that achieves commercial-grade model performance. We provide complete details of our training recipe, which utilizes an improved version of the SOTA SecAlign defense. Evaluations on 9 utility benchmarks and 7 security benchmarks show that Meta SecAlign, despite being trained on a generic instruction-tuning dataset, confers security in unseen downstream tasks, including tool-calling and agentic web navigation, in addition general instruction-following. Our best model – Meta-SecAlign-70B – achieves state-of-the-art robustness against prompt injection attacks and comparable utility to closed-source commercial LLM with model-level defense.

Large language models (LLMs) have seen widespread success in code generation tasks for different scenarios, both everyday and professional. However current LLMs, despite producing functional code, do not prioritize security and may generate code with exploitable vulnerabilities. In this work, we propose techniques for generating code that is more likely to be secure and introduce SCGAgent, a proactive secure coding agent that implements our techniques. We use security coding guidelines that articulate safe programming practices, combined with LLM-generated unit tests to preserve functional correctness. In our evaluation, we find that SCGAgent is able to preserve nearly 98 functionality of the base Sonnet-3.7 LLM while achieving an approximately 25 improvement in security. Moreover, SCGAgent is able to match or best the performance of sophisticated reasoning LLMs using a non-reasoning model and an agentic workflow.

Safety and security remain critical concerns in AI deployment. Despite safety training through reinforcement learning with human feedback (RLHF) [34], language models remain vulnerable to jailbreak attacks that bypass safety guardrails. Universal jailbreaks-prefixes that can circumvent alignment for any payload-are particularly concerning. We show empirically that jailbreak detection systems face distribution shift, with detectors trained at one point in time performing poorly against newer exploits. To study this problem, we release JailbreaksOverTime, a comprehensive dataset of timestamped real user interactions that contains benign requests and jailbreak attempts over the span of 10 months. We propose a two-pronged method for defenders to detect new jailbreaks and continuously update their detectors. First, we show how to use continuous learning to detect jailbreaks and adapt rapidly to new emerging jailbreaks. While detectors trained at a single point in time eventually fail due to drift, we find that universal jailbreaks evolve slowly enough for self-training to be effective. Retraining our detection model weekly using its own labels-with no new human labels-reduces the false negative rate from 4% to 0.3% at a false positive rate of 0.1%. Second, we introduce an unsupervised active monitoring approach to identify novel jailbreaks. Rather than classifying inputs directly, we recognize jailbreaks by their behavior, specifically, their ability to trigger models to respond to known-harmful prompts. We find that active monitoring can identify out-of-distribution attacks that were missed by continuous learning; overall, these methods provide insights on building jailbreak detectors that are resistant to concept drift.

When large language model (LLM) systems interact with external data to perform complex tasks, a new attack, namely prompt injection, becomes a significant threat. By injecting instructions into the data accessed by the system, the attacker is able to override the initial user task with an arbitrary task directed by the attacker. To secure the system, test-time defenses, e.g., defensive prompting, have been proposed for system developers to attain security only when needed in a flexible manner. However, they are much less effective than training-time defenses that change the model parameters. Motivated by this, we propose DefensiveToken, a test-time defense with prompt injection robustness comparable to training-time alternatives. DefensiveTokens are newly inserted as special tokens, whose embeddings are optimized for security. In security-sensitive cases, system developers can append a few DefensiveTokens before the LLM input to achieve security with a minimal utility drop. In scenarios where security is less of a concern, developers can simply skip DefensiveTokens; the LLM system remains the same as there is no defense, generating high-quality responses. Thus, DefensiveTokens, if released alongside the model, allow a flexible switch between the state-of-the-art (SOTA) utility and almost-SOTA security at test time. The code is available here.

We introduce SeedAIchemy, an automated LLM-driven corpus generation tool that makes it easier for developers to implement fuzzing effectively. SeedAIchemy consists of five modules which implement different approaches at collecting publicly available files from the internet. Four of the five modules use large language model (LLM) workflows to construct search terms designed to maximize corpus quality. Corpora generated by SeedAIchemy perform significantly better than a naive corpus and similarly to a manually-curated corpus on a diverse range of target programs and libraries.

When large language model (LLM) agents are increasingly deployed to automate tasks and interact with untrusted external data, prompt injection emerges as a significant security threat. By injecting malicious instructions into the data that LLMs access, an attacker can arbitrarily override the original user task and redirect the agent toward unintended, potentially harmful actions. Existing defenses either require access to model weights (fine-tuning), incur substantial utility loss (detection-based), or demand non-trivial system redesign (system-level). Motivated by this, we propose DataFilter, a test-time model-agnostic defense that removes malicious instructions from the data before it reaches the backend LLM. DataFilter is trained with supervised fine-tuning on simulated injections and leverages both the user's instruction and the data to selectively strip adversarial content while preserving benign information. Across multiple benchmarks, DataFilter consistently reduces the prompt injection attack success rates to near zero while maintaining the LLMs' utility. DataFilter delivers strong security, high utility, and plug-and-play deployment, making it a strong practical defense to secure black-box commercial LLMs against prompt injection. Our DataFilter model is released at https://huggingface.co/JoyYizhu/DataFilter for immediate use, with the code to reproduce our results at https://github.com/yizhu-joy/DataFilter.

Large language models (LLMs) are becoming increasingly prevalent in modern software systems, interfacing between the user and the Internet to assist with tasks that require advanced language understanding. To accomplish these tasks, the LLM often uses external data sources such as user documents, web retrieval, results from API calls, etc. This opens up new avenues for attackers to manipulate the LLM via prompt injection. Adversarial prompts can be injected into external data sources to override the system's intended instruction and instead execute a malicious instruction. To mitigate this vulnerability, we propose a new defense called SecAlign based on the technique of preference optimization. Our defense first constructs a preference dataset with prompt-injected inputs, secure outputs (ones that respond to the legitimate instruction), and insecure outputs (ones that respond to the injection). We then perform preference optimization on this dataset to teach the LLM to prefer the secure output over the insecure one. This provides the first known method that reduces the success rates of various prompt injections to <10%, even against attacks much more sophisticated than ones seen during training. This indicates our defense generalizes well against unknown and yet-to-come attacks. Also, SecAlign models are still practical with similar utility to the one before defensive training in our evaluations. Our code is here.

Application designers have moved to integrate large language models (LLMs) into their products. However, many LLM-integrated applications are vulnerable to prompt injections. While attempts have been made to address this problem by building prompt injection detectors, many are not yet suitable for practical deployment. To support research in this area, we introduce PromptShield, a benchmark for training and evaluating deployable prompt injection detectors. Our benchmark is carefully curated and includes both conversational and application-structured data. In addition, we use insights from our curation process to fine-tune a new prompt injection detector that achieves significantly higher performance in the low false positive rate (FPR) evaluation regime compared to prior schemes. Our work suggests that careful curation of training data and larger models can contribute to strong detector performance.

In the context of the rising interest in code language models (code LMs) and vulnerability detection, we study the effectiveness of code LMs for detecting vulnerabilities. Our analysis reveals significant shortcomings in existing vulnerability datasets, including poor data quality, low label accuracy, and high duplication rates, leading to unreliable model performance in realistic vulnerability detection scenarios. Additionally, the evaluation methods used with these datasets are not representative of real-world vulnerability detection. To address these challenges, we introduce PrimeVul, a new dataset for training and evaluating code LMs for vulnerability detection. PrimeVul incorporates a novel set of data labeling techniques that achieve comparable label accuracy to human-verified benchmarks while significantly expanding the dataset. It also implements a rigorous data de-duplication and chronological data splitting strategy to mitigate data leakage issues, alongside introducing more realistic evaluation metrics and settings. This comprehensive approach aims to provide a more accurate assessment of code LMs' performance in real-world conditions. Evaluating code LMs on PrimeVul reveals that existing benchmarks significantly overestimate the performance of these models. For instance, a state-of-the-art 7B model scored 68.26% F1 on BigVul but only 3.09% F1 on PrimeVul. Attempts to improve performance through advanced training techniques and larger models like GPT-3.5 and GPT-4 were unsuccessful, with results akin to random guessing in the most stringent settings. These findings underscore the considerable gap between current capabilities and the practical requirements for deploying code LMs in security roles, highlighting the need for more innovative research in this domain.