World models -- learned internal simulators of environment dynamics -- are rapidly becoming foundational to autonomous decision-making in robotics, autonomous vehicles, and agentic AI. Yet this predictive power introduces a distinctive set of safety, security, and cognitive risks. Adversaries can corrupt training data, poison latent representations, and exploit compounding rollout errors to cause catastrophic failures in safety-critical deployments. World model-equipped agents are more capable of goal misgeneralisation, deceptive alignment, and reward hacking precisely because they can simulate the consequences of their own actions. Authoritative world model predictions further foster automation bias and miscalibrated human trust that operators lack the tools to audit. This paper surveys the world model landscape; introduces formal definitions of trajectory persistence and representational risk; presents a five-profile attacker capability taxonomy; and develops a unified threat model extending MITRE ATLAS and the OWASP LLM Top 10 to the world model stack. We provide an empirical proof-of-concept on trajectory-persistent adversarial attacks (GRU-RSSM: A_1 = 2.26x amplification, -59.5% reduction under adversarial fine-tuning; stochastic RSSM proxy: A_1 = 0.65x; DreamerV3 checkpoint: non-zero action drift confirmed). We illustrate risks through four deployment scenarios and propose interdisciplinary mitigations spanning adversarial hardening, alignment engineering, NIST AI RMF and EU AI Act governance, and human-factors design. We argue that world models must be treated as safety-critical infrastructure requiring the same rigour as flight-control software or medical devices.

Personal AI agents like OpenClaw run with elevated privileges on users' local machines, where a single successful prompt injection can leak credentials, redirect financial transactions, or destroy files. This threat goes well beyond conventional text-level jailbreaks, yet existing safety evaluations fall short: most test models in isolated chat settings, rely on synthetic environments, and do not account for how the agent framework itself shapes safety outcomes. We introduce CLAWSAFETY, a benchmark of 120 adversarial test scenarios organized along three dimensions (harm domain, attack vector, and harmful action type) and grounded in realistic, high-privilege professional workspaces spanning software engineering, finance, healthcare, law, and DevOps. Each test case embeds adversarial content in one of three channels the agent encounters during normal work: workspace skill files, emails from trusted senders, and web pages. We evaluate five frontier LLMs as agent backbones, running 2,520 sandboxed trials across all configurations. Attack success rates (ASR) range from 40\% to 75\% across models and vary sharply by injection vector, with skill instructions (highest trust) consistently more dangerous than email or web content. Action-trace analysis reveals that the strongest model maintains hard boundaries against credential forwarding and destructive actions, while weaker models permit both. Cross-scaffold experiments on three agent frameworks further demonstrate that safety is not determined by the backbone model alone but depends on the full deployment stack, calling for safety evaluation that treats model and framework as joint variables.

Large language models (LLMs) and their applications, such as agents, are highly vulnerable to prompt injection attacks. State-of-the-art prompt injection detection methods have the following limitations: (1) their effectiveness degrades significantly as context length increases, and (2) they lack explicit rules that define what constitutes prompt injection, causing detection decisions to be implicit, opaque, and difficult to reason about. In this work, we propose AgentWatcher to address the above two limitations. To address the first limitation, AgentWatcher attributes the LLM's output (e.g., the action of an agent) to a small set of causally influential context segments. By focusing detection on a relatively short text, AgentWatcher can be scalable to long contexts. To address the second limitation, we define a set of rules specifying what does and does not constitute a prompt injection, and use a monitor LLM to reason over these rules based on the attributed text, making the detection decisions more explainable. We conduct a comprehensive evaluation on tool-use agent benchmarks and long-context understanding datasets. The experimental results demonstrate that AgentWatcher can effectively detect prompt injection and maintain utility without attacks. The code is available at https://github.com/wang-yanting/AgentWatcher.

Autonomous AI agents are being deployed with filesystem access, email control, and multi-step planning. This thesis contributes to four open problems in AI safety: understanding dangerous internal computations, removing dangerous behaviors once embedded, testing for vulnerabilities before deployment, and predicting when models will act against deployers. ACDC automates circuit discovery in transformers, recovering all five component types from prior manual work on GPT-2 Small by selecting 68 edges from 32,000 candidates in hours rather than months. Latent Adversarial Training (LAT) removes dangerous behaviors by optimizing perturbations in the residual stream to elicit failure modes, then training under those perturbations. LAT solved the sleeper agent problem where standard safety training failed, matching existing defenses with 700x fewer GPU hours. Best-of-N jailbreaking achieves 89% attack success on GPT-4o and 78% on Claude 3.5 Sonnet through random input augmentations. Attack success follows power law scaling across text, vision, and audio, enabling quantitative forecasting of adversarial robustness. Agentic misalignment tests whether frontier models autonomously choose harmful actions given ordinary goals. Across 16 models, agents engaged in blackmail (96% for Claude Opus 4), espionage, and actions causing death. Misbehavior rates rose from 6.5% to 55.1% when models stated scenarios were real rather than evaluations. The thesis does not fully resolve any of these problems but makes each tractable and measurable.

Equipping LLM agents with real-world tools can substantially improve productivity. However, granting agents autonomy over tool use also transfers the associated privileges to both the agent and the underlying LLM. Improper privilege usage may lead to serious consequences, including information leakage and infrastructure damage. While several benchmarks have been built to study agents' security, they often rely on pre-coded tools and restricted interaction patterns. Such crafted environments differ substantially from the real-world, making it hard to assess agents' security capabilities in critical privilege control and usage. Therefore, we propose GrantBox, a security evaluation sandbox for analyzing agent privilege usage. GrantBox automatically integrates real-world tools and allows LLM agents to invoke genuine privileges, enabling the evaluation of privilege usage under prompt injection attacks. Our results indicate that while LLMs exhibit basic security awareness and can block some direct attacks, they remain vulnerable to more sophisticated attacks, resulting in an average attack success rate of 84.80% in carefully crafted scenarios.

System prompt configuration can make the difference between near-total phishing blindness and near-perfect detection in LLM email agents. We present PhishNChips, a study of 11 models under 10 prompt strategies, showing that prompt-model interaction is a first-order security variable: a single model's phishing bypass rate ranges from under 1% to 97% depending on how it is configured, while the false-positive cost of the same prompt varies sharply across models. We then show that optimizing prompts around highly predictive signals can improve benchmark performance, reaching up to 93.7% recall at 3.8% false positive rate, but also creates a brittle attack surface. In particular, domain-matching strategies perform well when legitimate emails mostly have matched sender and URL domains, yet degrade sharply when attackers invert that signal by registering matching infrastructure. Response-trace analysis shows that 98% of successful bypasses reason in ways consistent with the inverted signal: the models are following the instruction, but the instruction's core assumption has become false. A counter-intuitive corollary follows: making prompts more specific can degrade already-capable models by replacing broader multi-signal reasoning with exploitable single-signal dependence. We characterize the resulting tension between detection, usability, and adversarial robustness as a navigable tradeoff, introduce Safetility, a deployability-aware metric that penalizes false positives, and argue that closing the adversarial gap likely requires tool augmentation with external ground truth.

Agentic computing systems, which autonomously spawn new functionalities based on natural language instructions, are becoming increasingly prevalent. While immensely capable, these systems raise serious security, privacy, and safety concerns. Fundamentally, the full set of functionalities offered by these systems, combined with their probabilistic execution flows, is not known beforehand. Given this lack of characterization, it is non-trivial to validate whether a system has successfully carried out the user's intended task or instead executed irrelevant actions, potentially as a consequence of compromise. In this paper, we propose Agent-Sentry, a framework that attempts to bound agentic systems to address this problem. Our key insight is that agentic systems are designed for specific use cases and therefore need not expose unbounded or unspecified functionalities. Once bounded, these systems become easier to scrutinize. Agent-Sentry operationalizes this insight by uncovering frequent functionalities offered by an agentic system, along with their execution traces, to construct behavioral bounds. It then learns a policy from these traces and blocks tool calls that deviate from learned behaviors or that misalign with user intent. Our evaluation shows that Agent-Sentry helps prevent over 90\% of attacks that attempt to trigger out-of-bounds executions, while preserving up to 98\% of system utility.

Recent AI systems combine large language models with tools, external knowledge via retrieval-augmented generation (RAG), and even autonomous multi-agent decision loops. This agentic AI paradigm greatly expands capabilities - but also vastly enlarges the attack surface. In this systematization, we map out the trust boundaries and security risks of agentic LLM-based systems. We develop a comprehensive taxonomy of attacks spanning prompt-level injections, knowledge-base poisoning, tool/plug-in exploits, and multi-agent emergent threats. Through a detailed literature review, we synthesize evidence from 2023-2025, including more than 20 peer-reviewed and archival studies, industry reports, and standards. We find that agentic systems introduce new vectors for indirect prompt injection, code execution exploits, RAG index poisoning, and cross-agent manipulation that go beyond traditional AI threats. We define attacker models and threat scenarios, and propose metrics (e.g., Unsafe Action Rate, Privilege Escalation Distance) to evaluate security posture. Our survey examines defenses such as input sanitization, retrieval filters, sandboxes, access control, and "AI guardrails," assessing their effectiveness and pointing out the areas where protection is still lacking. To assist practitioners, we outline defensive controls and provide a phased security checklist for deploying agentic AI (covering design-time hardening, runtime monitoring, and incident response). Finally, we outline open research challenges in secure autonomous AI (robust tool APIs, verifiable agent behavior, supply-chain safeguards) and discuss ethical and responsible disclosure practices. We systematize recent findings to help researchers and engineers understand and mitigate security risks in agentic AI.

An autonomous AI ecosystem (SUBSTRATE S3), generating product specifications without explicit instructions about formal methods, independently proposed the use of Z3 SMT solver across six distinct domains of AI safety: verification of LLM-generated code, tool API safety for AI agents, post-distillation reasoning correctness, CLI command validation, hardware assembly verification, and smart contract safety. These convergent discoveries, occurring across 8 products over 13 days with Jaccard similarity below 15% between variants, suggest that formal verification is not merely a useful technique for AI safety but an emergent property of any sufficiently complex system reasoning about its own safety. We propose a unified framework (substrate-guard) that applies Z3-based verification across all six output classes through a common API, and evaluate it on 181 test cases across five implemented domains, achieving 100% classification accuracy with zero false positives and zero false negatives. Our framework detected real bugs that empirical testing would miss, including an INT_MIN overflow in branchless RISC-V assembly and mathematically proved that unconstrained string parameters in tool APIs are formally unverifiable.

Multi-agent discussions have been widely adopted, motivating growing efforts to develop attacks that expose their vulnerabilities. In this work, we study a practical yet largely unexplored attack scenario, the discussion-monitored scenario, where anomaly detectors continuously monitor inter-agent communications and block detected adversarial messages. Although existing attacks are effective without discussion monitoring, we show that they exhibit detectable patterns and largely fail under such monitoring constraints. But does this imply that monitoring alone is sufficient to secure multi-agent discussions? To answer this question, we develop a novel attack method explicitly tailored to the discussion-monitored scenario. Extensive experiments demonstrate that effective attacks remain possible even under continuous monitoring, indicating that monitoring alone does not eliminate adversarial risks.

LLM agent frameworks increasingly offer checkpoint-restore for error recovery and exploration, advising developers to make external tool calls safe to retry. This advice assumes that a retried call will be identical to the original, an assumption that holds for traditional programs but fails for LLM agents, which re-synthesize subtly different requests after restore. Servers treat these re-generated requests as new, enabling duplicate payments, unauthorized reuse of consumed credentials, and other irreversible side effects; we term these semantic rollback attacks. We identify two attack classes, Action Replay and Authority Resurrection, validate them in a proof of concept experiment, and confirm that the problem has been independently acknowledged by framework maintainers. We propose ACRFence, a framework-agnostic mitigation that records irreversible tool effects and enforces replay-or-fork semantics upon restoration

Memory poisoning attacks for Agentic AI and multi-agent systems (MAS) have recently caught attention. It is partially due to the fact that Large Language Models (LLMs) facilitate the construction and deployment of agents. Different memory systems are being used nowadays in this context, including semantic, episodic, and short-term memory. This distinction between the different types of memory systems focuses mostly on their duration but also on their origin and their localization. It ranges from the short-term memory originated at the user's end localized in the different agents to the long-term consolidated memory localized in well established knowledge databases. In this paper, we first present the main types of memory systems, we then discuss the feasibility of memory poisoning attacks in these different types of memory systems, and we propose mitigation strategies. We review the already existing security solutions to mitigate some of the alleged attacks, and we discuss adapted solutions based on cryptography. We propose to implement local inference based on private knowledge retrieval as an example of mitigation strategy for memory poisoning for semantic memory. We also emphasize actual risks in relation to interactions between agents, which can cause memory poisoning. These latter risks are not so much studied in the literature and are difficult to formalize and solve. Thus, we contribute to the construction of agents that are secure by design.

Large language model (LLM) agents increasingly rely on external tools (file operations, API calls, database transactions) to autonomously complete complex multi-step tasks. Practitioners deploy defense-trained models to protect against prompt injection attacks that manipulate agent behavior through malicious observations or retrieved content. We reveal a fundamental \textbf{capability-alignment paradox}: defense training designed to improve safety systematically destroys agent competence while failing to prevent sophisticated attacks. Evaluating defended models against undefended baselines across 97 agent tasks and 1,000 adversarial prompts, we uncover three systematic biases unique to multi-step agents. \textbf{Agent incompetence bias} manifests as immediate tool execution breakdown, with models refusing or generating invalid actions on benign tasks before observing any external content. \textbf{Cascade amplification bias} causes early failures to propagate through retry loops, pushing defended models to timeout on 99\% of tasks compared to 13\% for baselines. \textbf{Trigger bias} leads to paradoxical security degradation where defended models perform worse than undefended baselines while straightforward attacks bypass defenses at high rates. Root cause analysis reveals these biases stem from shortcut learning: models overfit to surface attack patterns rather than semantic threat understanding, evidenced by extreme variance in defense effectiveness across attack categories. Our findings demonstrate that current defense paradigms optimize for single-turn refusal benchmarks while rendering multi-step agents fundamentally unreliable, necessitating new approaches that preserve tool execution competence under adversarial conditions.

Autonomous web agents such as \textbf{OpenClaw} are rapidly moving into high-impact real-world workflows, but their security robustness under live network threats remains insufficiently evaluated. Existing benchmarks mainly focus on static sandbox settings and content-level prompt attacks, which leaves a practical gap for network-layer security testing. In this paper, we present \textbf{ClawTrap}, a \textbf{MITM-based red-teaming framework for real-world OpenClaw security evaluation}. ClawTrap supports diverse and customizable attack forms, including \textit{Static HTML Replacement}, \textit{Iframe Popup Injection}, and \textit{Dynamic Content Modification}, and provides a reproducible pipeline for rule-driven interception, transformation, and auditing. This design lays the foundation for future research to construct richer, customizable MITM attacks and to perform systematic security testing across agent frameworks and model backbones. Our empirical study shows clear model stratification: weaker models are more likely to trust tampered observations and produce unsafe outputs, while stronger models demonstrate better anomaly attribution and safer fallback strategies. These findings indicate that reliable OpenClaw security evaluation should explicitly incorporate dynamic real-world MITM conditions rather than relying only on static sandbox protocols.

Agentic LLM systems equipped with persistent memory, RAG pipelines, and external tool connectors face a class of attacks - Logic-layer Prompt Control Injection (LPCI) - for which no automated red-teaming instrument existed. We present LAAF (Logic-layer Automated Attack Framework), the first automated red-teaming framework to combine an LPCI-specific technique taxonomy with stage-sequential seed escalation - two capabilities absent from existing tools: Garak lacks memory-persistence and cross-session triggering; PyRIT supports multi-turn testing but treats turns independently, without seeding each stage from the prior breakthrough. LAAF provides: (i) a 49-technique taxonomy spanning six attack categories (Encoding~11, Structural~8, Semantic~8, Layered~5, Trigger~12, Exfiltration~5; see Table 1), combinable across 5 variants per technique and 6 lifecycle stages, yielding a theoretical maximum of 2,822,400 unique payloads ($49 \times 5 \times 1{,}920 \times 6$; SHA-256 deduplicated at generation time); and (ii) a Persistent Stage Breaker (PSB) that drives payload mutation stage-by-stage: on each breakthrough, the PSB seeds the next stage with a mutated form of the winning payload, mirroring real adversarial escalation. Evaluation on five production LLM platforms across three independent runs demonstrates that LAAF achieves higher stage-breakthrough efficiency than single-technique random testing, with a mean aggregate breakthrough rate of 84\% (range 83--86\%) and platform-level rates stable within 17 percentage points across runs. Layered combinations and semantic reframing are the highest-effectiveness technique categories, with layered payloads outperforming encoding on well-defended platforms.

Agentic AI has been a topic of great interest recently. A Large Language Model (LLM) agent involves one or more LLMs in the back-end. In the front end, it conducts autonomous decision-making by combining the LLM outputs with results obtained by invoking several external tools. The autonomous interactions with the external environment introduce critical security risks. In this paper, we present a grey-box approach to explore diverse behaviors and uncover security risks in LLM agents. Our approach VeriGrey uses the sequence of tools invoked as a feedback function to drive the testing process. This helps uncover infrequent but dangerous tool invocations that cause unexpected agent behavior. As mutation operators in the testing process, we mutate prompts to design pernicious injection prompts. This is carefully accomplished by linking the task of the agent to an injection task, so that the injection task becomes a necessary step of completing the agent functionality. Comparing our approach with a black-box baseline on the well-known AgentDojo benchmark, VeriGrey achieves 33% additional efficacy in finding indirect prompt injection vulnerabilities with a GPT-4.1 back-end. We also conduct real-world case studies with the widely used coding agent Gemini CLI, and the well-known OpenClaw personal assistant. VeriGrey finds prompts inducing several attack scenarios that could not be identified by black-box approaches. In OpenClaw, by constructing a conversation agent which employs mutational fuzz testing as needed, VeriGrey is able to discover malicious skill variants from 10 malicious skills (with 10/10= 100% success rate on the Kimi-K2.5 LLM backend, and 9/10= 90% success rate on Opus 4.6 LLM backend). This demonstrates the value of a dynamic approach like VeriGrey to test agents, and to eventually lead to an agent assurance framework.

Autonomous AI agents powered by large language models are being deployed in production with capabilities including shell execution, file system access, database queries, and multi-party communication. Recent red teaming research demonstrates that these agents exhibit critical vulnerabilities in realistic settings: unauthorized compliance with non-owner instructions, sensitive information disclosure, identity spoofing, cross-agent propagation of unsafe practices, and indirect prompt injection through external resources [7]. In healthcare environments processing Protected Health Information, every such vulnerability becomes a potential HIPAA violation. This paper presents a security architecture deployed for nine autonomous AI agents in production at a healthcare technology company. We develop a six-domain threat model for agentic AI in healthcare covering credential exposure, execution capability abuse, network egress exfiltration, prompt integrity failures, database access risks, and fleet configuration drift. We implement four-layer defense in depth: (1) kernel level workload isolation using gVisor on Kubernetes, (2) credential proxy sidecars preventing agent containers from accessing raw secrets, (3) network egress policies restricting each agent to allowlisted destinations, and (4) a prompt integrity framework with structured metadata envelopes and untrusted content labeling. We report results from 90 days of deployment including four HIGH severity findings discovered and remediated by an automated security audit agent, progressive fleet hardening across three VM image generations, and defense coverage mapped to all eleven attack patterns from recent literature. All configurations, audit tooling, and the prompt integrity framework are released as open source.

LLM based agents are increasingly deployed in high stakes settings where they process external data sources such as emails, documents, and code repositories. This creates exposure to indirect prompt injection attacks, where adversarial instructions embedded in external content manipulate agent behavior without user awareness. A critical but underexplored dimension of this threat is concealment: since users tend to observe only an agent's final response, an attack can conceal its existence by presenting no clue of compromise in the final user facing response while successfully executing harmful actions. This leaves users unaware of the manipulation and likely to accept harmful outcomes as legitimate. We present findings from a large scale public red teaming competition evaluating this dual objective across three agent settings: tool calling, coding, and computer use. The competition attracted 464 participants who submitted 272000 attack attempts against 13 frontier models, yielding 8648 successful attacks across 41 scenarios. All models proved vulnerable, with attack success rates ranging from 0.5% (Claude Opus 4.5) to 8.5% (Gemini 2.5 Pro). We identify universal attack strategies that transfer across 21 of 41 behaviors and multiple model families, suggesting fundamental weaknesses in instruction following architectures. Capability and robustness showed weak correlation, with Gemini 2.5 Pro exhibiting both high capability and high vulnerability. To address benchmark saturation and obsoleteness, we will endeavor to deliver quarterly updates through continued red teaming competitions. We open source the competition environment for use in evaluations, along with 95 successful attacks against Qwen that did not transfer to any closed source model. We share model-specific attack data with respective frontier labs and the full dataset with the UK AISI and US CAISI to support robustness research.

With the rapid development of LLM-based multi-agent systems (MAS), their significant safety and security concerns have emerged, which introduce novel risks going beyond single agents or LLMs. Despite attempts to address these issues, the existing literature lacks a cohesive safeguarding system specialized for MAS risks. In this work, we introduce TrinityGuard, a comprehensive safety evaluation and monitoring framework for LLM-based MAS, grounded in the OWASP standards. Specifically, TrinityGuard encompasses a three-tier fine-grained risk taxonomy that identifies 20 risk types, covering single-agent vulnerabilities, inter-agent communication threats, and system-level emergent hazards. Designed for scalability across various MAS structures and platforms, TrinityGuard is organized in a trinity manner, involving an MAS abstraction layer that can be adapted to any MAS structures, an evaluation layer containing risk-specific test modules, alongside runtime monitor agents coordinated by a unified LLM Judge Factory. During Evaluation, TrinityGuard executes curated attack probes to generate detailed vulnerability reports for each risk type, where monitor agents analyze structured execution traces and issue real-time alerts, enabling both pre-development evaluation and runtime monitoring. We further formalize these safety metrics and present detailed case studies across various representative MAS examples, showcasing the versatility and reliability of TrinityGuard. Overall, TrinityGuard acts as a comprehensive framework for evaluating and monitoring various risks in MAS, paving the way for further research into their safety and security.

Modern agentic systems allow Large Language Model (LLM) agents to tackle complex tasks through extensive tool usage, forming structured control flows of tool selection and execution. Existing security analyses often treat these control flows as ephemeral, one-off sessions, overlooking the persistent influence of memory. This paper identifies a new threat from Memory Control Flow Attacks (MCFA) that memory retrieval can dominate the control flow, forcing unintended tool usage even against explicit user instructions and inducing persistent behavioral deviations across tasks. To understand the impact of this vulnerability, we further design MEMFLOW, an automated evaluation framework that systematically identifies and quantifies MCFA across heterogeneous tasks and long interaction horizons. To evaluate MEMFLOW, we attack state-of-the-art LLMs, including GPT-5 mini, Claude Sonnet 4.5 and Gemini 2.5 Flash on real-world tools from two major LLM agent development frameworks, LangChain and LlamaIndex. The results show that in general over 90% trials are vulnerable to MCFA even under strict safety constraints, highlighting critical security risks that demand immediate attention.