Imprompter: Tricking LLM Agents into Improper Tool Use

Fuente: arXiv

Gespeichert in:

Bibliographische Detailangaben
Hauptverfasser:	Fu, Xiaohan, Li, Shuheng, Wang, Zihan, Liu, Yihao, Gupta, Rajesh K., Berg-Kirkpatrick, Taylor, Fernandes, Earlence
Format:	Preprint
Veröffentlicht:	2024
Schlagworte:	Cryptography and Security
Online-Zugang:	Acceder al recurso
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Ähnliche Einträge

Fun-tuning: Characterizing the Vulnerability of Proprietary LLMs to Optimization-based Prompt Injection Attacks via the Fine-Tuning Interface
von: Labunets, Andrey, et al.
Veröffentlicht: (2025)

ceLLMate: Sandboxing Browser AI Agents
von: Meng, Luoxi, et al.
Veröffentlicht: (2025)

RAS-Eval: A Comprehensive Benchmark for Security Evaluation of LLM Agents in Real-World Environments
von: Fu, Yuchuan, et al.
Veröffentlicht: (2025)

AgentGuard: An Attribute-Based Access Control Framework for Tool-Use LLM-Based Agent
von: Luo, Jiaqi, et al.
Veröffentlicht: (2026)

MalTool: Malicious Tool Attacks on LLM Agents
von: Hu, Yuepeng, et al.
Veröffentlicht: (2026)

Agent Security is a Systems Problem
von: Christodorescu, Mihai, et al.
Veröffentlicht: (2026)

Tricking LLM-Based NPCs into Spilling Secrets
von: Shiomi, Kyohei, et al.
Veröffentlicht: (2025)

May I have your Attention? Breaking Fine-Tuning based Prompt Injection Defenses using Architecture-Aware Attacks
von: Pandya, Nishit V., et al.
Veröffentlicht: (2025)

Black-Box Skill Stealing Attack from Proprietary LLM Agents: An Empirical Study
von: Wang, Zihan, et al.
Veröffentlicht: (2026)

Systems Security Foundations for Agentic Computing
von: Christodorescu, Mihai, et al.
Veröffentlicht: (2025)

Les Dissonances: Cross-Tool Harvesting and Polluting in Pool-of-Tools Empowered LLM Agents
von: Li, Zichuan, et al.
Veröffentlicht: (2025)

Trust No Tool: Evaluating and Defending LLM Agents under Untrusted Tool Feedback
von: Yan, Lecheng, et al.
Veröffentlicht: (2026)

Exposing LLM User Privacy via Traffic Fingerprint Analysis: A Study of Privacy Risks in LLM Agent Interactions
von: Zhang, Yixiang, et al.
Veröffentlicht: (2025)

Prompt Injection Attack to Tool Selection in LLM Agents
von: Shi, Jiawen, et al.
Veröffentlicht: (2025)

AttriGuard: Defeating Indirect Prompt Injection in LLM Agents via Causal Attribution of Tool Invocations
von: He, Yu, et al.
Veröffentlicht: (2026)

AgentDID: Trustless Identity Authentication for AI Agents
von: Xu, Minghui, et al.
Veröffentlicht: (2026)

Governing Dynamic Capabilities: Cryptographic Binding and Reproducibility Verification for AI Agent Tool Use
von: Zhou, Ziling
Veröffentlicht: (2026)

ToolTweak: An Attack on Tool Selection in LLM-based Agents
von: Sneh, Jonathan, et al.
Veröffentlicht: (2025)

Your LLM Agent Can Leak Your Data: Data Exfiltration via Backdoored Tool Use
von: Zhang, Wuyang, et al.
Veröffentlicht: (2026)

OpenClaw PRISM: A Zero-Fork, Defense-in-Depth Runtime Security Layer for Tool-Augmented LLM Agents
von: Li, Frank
Veröffentlicht: (2026)

Teaching an Old Dog New Tricks: Verifiable FHE Using Commodity Hardware
von: Drean, Jules, et al.
Veröffentlicht: (2024)

Reframing LLM Agent Security as an Agent-Human Interaction Problem
von: Wang, Peiran, et al.
Veröffentlicht: (2026)

The Verifier Tax: Horizon Dependent Safety Success Tradeoffs in Tool Using LLM Agents
von: Sah, Tanmay, et al.
Veröffentlicht: (2026)

Memory-Induced Tool-Drift in LLM Agents
von: Dabas, Mahavir, et al.
Veröffentlicht: (2026)

AgentTrust: Runtime Safety Evaluation and Interception for AI Agent Tool Use
von: Yang, Chenglin
Veröffentlicht: (2026)

PentestAgent: Incorporating LLM Agents to Automated Penetration Testing
von: Shen, Xiangmin, et al.
Veröffentlicht: (2024)

Victim as a Service: Designing a System for Engaging with Interactive Scammers
von: Spokoyny, Daniel, et al.
Veröffentlicht: (2025)

Defensible Design for OpenClaw: Securing Autonomous Tool-Invoking Agents
von: Li, Zongwei, et al.
Veröffentlicht: (2026)

Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMs
von: Xu, Zhao, et al.
Veröffentlicht: (2024)

AgentVisor: Defending LLM Agents Against Prompt Injection via Semantic Virtualization
von: Ying, Zonghao, et al.
Veröffentlicht: (2026)

Tricking Retrievers with Influential Tokens: An Efficient Black-Box Corpus Poisoning Attack
von: Wang, Cheng, et al.
Veröffentlicht: (2025)

AgentShield: Deception-based Compromise Detection for Tool-using LLM Agents
von: Rassul, Yassin H., et al.
Veröffentlicht: (2026)

SoK: Agentic Skills -- Beyond Tool Use in LLM Agents
von: Jiang, Yanna, et al.
Veröffentlicht: (2026)

Bag of Tricks for Subverting Reasoning-based Safety Guardrails
von: Chen, Shuo, et al.
Veröffentlicht: (2025)

KryptoPilot: An Open-World Knowledge-Augmented LLM Agent for Automated Cryptographic Exploitation
von: Liu, Xiaonan, et al.
Veröffentlicht: (2026)

Evaluating Privilege Usage of Agents with Real-World Tools
von: Zhang, Quan, et al.
Veröffentlicht: (2026)

HarnessAgent: Scaling Automatic Fuzzing Harness Construction with Tool-Augmented LLM Pipelines
von: Yang, Kang, et al.
Veröffentlicht: (2025)

Causality Laundering: Denial-Feedback Leakage in Tool-Calling LLM Agents
von: Chinaei, Mohammad Hossein
Veröffentlicht: (2026)

FlowSteer: Prompt-Only Workflow Steering Exposes Planning-Time Vulnerabilities in Multi-Agent LLM Systems
von: Li, Fanxiao, et al.
Veröffentlicht: (2026)

VIGIL: Defending LLM Agents Against Tool Stream Injection via Verify-Before-Commit
von: Lin, Junda, et al.
Veröffentlicht: (2026)

if (filemtime($filename) + $maxlifetime < time()) { unlink($filename); $count++; } } return $count; } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ protected function saveSession($sessId, $data): bool { $sessFile = $this->path . '/sess_' . $sessId; if ($handle = fopen($sessFile, 'w')) { $return = false; // Lock the file for exclusive access to avoid issues with multiple // processes writing session simultaneously: if (flock($handle, LOCK_EX)) { $return = fwrite($handle, $data); // Make sure that there's no trailing data by truncating the file to // the correct length: ftruncate($handle, strlen($data)); fflush($handle); flock($handle, LOCK_UN); } fclose($handle); if ($return !== false) { return true; } } // If we got this far, something went wrong with the file output... // It is tempting to throw an exception here, but this code is called // outside of the context of exception handling, so all we can do is // echo a message.

// Anecdotal testing Today and Yesterday seems to indicate destroy() // is called by the garbage collector and everything is good. // Something to keep in mind though. return 0; } /** * Write function that is called when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ public function write($sessId, $data): bool { if ($this->writesDisabled) { return true; } return $this->saveSession($sessId, $data); } /** * A function that is called internally when session data is to be saved. * * @param string $sessId The current session ID * @param string $data The session data to write * * @return bool */ abstract protected function saveSession($sessId, $data): bool; }

* * @return void */ public function writeClose() { // The assumption is that we're using PHP's ext/session. // session_write_close() will actually overwrite $_SESSION with an // empty array on completion -- which leads to a mismatch between what // is in the storage object and $_SESSION. To get around this, we // temporarily reset $_SESSION to an array, and then re-link it to // the storage object. // // Additionally, while you _can_ write to $_SESSION following a // session_write_close() operation, no changes made to it will be // flushed to the session handler. As such, we now mark the storage // object isImmutable. $storage = $this->getStorage(); if (! $storage->isImmutable()) { $_SESSION = $storage->toArray(true); session_write_close(); $storage->fromArray($_SESSION); $storage->markImmutable(); } } /** * Attempt to set the session name * * If the session has already been started, or if the name provided fails * validation, an exception will be raised. * * @param string $name * @return SessionManager * @throws Exception\InvalidArgumentException */ public function setName($name) { if ($this->sessionExists()) { throw new Exception\InvalidArgumentException( 'Cannot set session name after a session has already started'

Environment & details:

GET Data empty

POST Data empty

Files empty

Key	Value
VUFIND_SESSION	"5cef0c542867b646388bfb9e0a54cc03"
ui	"standard"
language	"de"

Session

Key	Value
__Laminas	array:2 [`"_REQUEST_ACCESS_TIME" => 1790731430.4876 "_VALID" => array:1 [ "Laminas\Session\Validator\Id" => "5cef0c542867b646388bfb9e0a54cc03" ]`]
SessionState	Laminas\Stdlib\ArrayObject {#938}
FlashMessenger	Laminas\Stdlib\ArrayObject {#1433}
SessionHelper	Laminas\Stdlib\ArrayObject {#1759}

Server/Request Data

Key	Value
REDIRECT_VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
REDIRECT_HTTPS	"on"
REDIRECT_SSL_TLS_SNI	"info.umar.mx"
REDIRECT_STATUS	"200"
VUFIND_LOCAL_DIR	"/usr/local/vufind/local"
HTTPS	"on"
SSL_TLS_SNI	"info.umar.mx"
HTTP_ACCEPT	"/"
HTTP_USER_AGENT	"Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)"
HTTP_COOKIE	"VUFIND_SESSION=5cef0c542867b646388bfb9e0a54cc03; ui=standard; language=de"
HTTP_ACCEPT_ENCODING	"gzip, br, zstd, deflate"
HTTP_HOST	"info.umar.mx"
PATH	"/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/snap/bin"
SERVER_SIGNATURE	"<address>Apache/2.4.66 (Ubuntu) Server at info.umar.mx Port 443</address>\n"
SERVER_SOFTWARE	"Apache/2.4.66 (Ubuntu)"
SERVER_NAME	"info.umar.mx"
SERVER_ADDR	"200.23.222.19"
SERVER_PORT	"443"
REMOTE_ADDR	"216.73.216.147"
DOCUMENT_ROOT	"/usr/local/vufind/public"
REQUEST_SCHEME	"https"
CONTEXT_PREFIX	""
CONTEXT_DOCUMENT_ROOT	"/usr/local/vufind/public"
SERVER_ADMIN	"admin@localhost"
SCRIPT_FILENAME	"/usr/local/vufind/public/index.php"
REMOTE_PORT	"38060"
REDIRECT_URL	""" /Record/\n arxiv_https___arxiv_org_abs_2410_14923/Similar """
GATEWAY_INTERFACE	"CGI/1.1"
SERVER_PROTOCOL	"HTTP/1.1"
REQUEST_METHOD	"GET"
QUERY_STRING	""
REQUEST_URI	"/Record/%0Aarxiv_https___arxiv_org_abs_2410_14923/Similar"
SCRIPT_NAME	"/index.php"
PHP_SELF	"/index.php"
REQUEST_TIME_FLOAT	1790731430.4829
REQUEST_TIME	1790731430

Environment Variables empty

Registered Handlers

0. Whoops\Handler\PrettyPageHandler