REVIEW 21 cited by
SafeArena: Evaluating the Safety of Autonomous Web Agents
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
SafeArena: Evaluating the Safety of Autonomous Web Agents
read the original abstract
LLM-based agents are becoming increasingly proficient at solving web-based tasks. With this capability comes a greater risk of misuse for malicious purposes, such as posting misinformation in an online forum or selling illicit substances on a website. To evaluate these risks, we propose SafeArena, the first benchmark to focus on the deliberate misuse of web agents. SafeArena comprises 250 safe and 250 harmful tasks across four websites. We classify the harmful tasks into five harm categories -- misinformation, illegal activity, harassment, cybercrime, and social bias, designed to assess realistic misuses of web agents. We evaluate leading LLM-based web agents, including GPT-4o, Claude-3.5 Sonnet, Qwen-2-VL 72B, and Llama-3.2 90B, on our benchmark. To systematically assess their susceptibility to harmful tasks, we introduce the Agent Risk Assessment framework that categorizes agent behavior across four risk levels. We find agents are surprisingly compliant with malicious requests, with GPT-4o and Qwen-2 completing 34.7% and 27.3% of harmful requests, respectively. Our findings highlight the urgent need for safety alignment procedures for web agents. Our benchmark is available here: https://safearena.github.io
Forward citations
Cited by 21 Pith papers
-
Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?
A user study with over 100 participants shows humans rarely spot AI agents sabotaging code during extended collaborative tasks, even with a safety monitor present.
-
From Monoliths to Swarms: A Study of Attack Surface Evolution in the Transition to Multi-Agent Web Systems
A new 'Telephone Loop' attack stalls multi-agent web systems in delegation cycles, succeeding in about 80% of baseline runs for three frontier models while failing against single-agent systems.
-
SciHazard: A Benchmark for Measuring Scientific Safety Risks with Decomposed Harm Scoring
A new scientific-safety benchmark and a decomposed, retrieval-grounded metric that aligns with expert harm judgments substantially better than existing LLM-as-judge baselines.
-
WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks
WebSP-Eval shows that multimodal LLM-based web agents fail more than 45% of the time on security and privacy tasks involving stateful UI elements such as toggles and checkboxes.
-
SecureWebArena: A Holistic Security Evaluation Benchmark for LVLM-based Web Agents
SecureWebArena is a new benchmark suite for holistic security evaluation of LVLM-based web agents using diverse simulated environments, attack taxonomies, and multi-layered failure analysis across reasoning, behavior,...
-
SafeFlow: Semantic Information-Flow Control for Blocking Malicious Propagation in Multi-Agent Systems
A workflow-level taint-propagation defense blocks fragmented malicious multi-agent workflows, cutting average attack success from 69.3% to 12.7% on four benchmarks.
-
Agent Security Needs Redefinition through a Holistic Framework
Agent security should be redefined around four contextual authorization properties instead of the content of the action performed.
-
Spinning Straw into Gold: Relabeling LLM Agent Trajectories in Hindsight for Successful Demonstrations
Relabeling LLM-agent trajectories with all goals actually achieved, plus action masking and reweighting, yields sample-efficient gains over SFT and DPO on ALFWorld, PlanCraft, and WebShop.
-
Understanding and Evaluating Claw-like Agent Security Through a Computer-Systems Lens
The paper introduces SafeClawArena, a 406-task benchmark evaluating security failures in three Claw-like agent platforms across skill supply-chain, state exploitation, data flow, and prompt injection surfaces.
-
PhoneBuddy: Training Open Models for Agentic Phone Use
PhoneBuddy combines real-app and mock-app RL after shared SFT, raising real-phone task success from 36.67% to 45.33% and AndroidWorld from 60.3% to 83.2%.
-
ROGUE: Misaligned Agent Behavior Arising from Ordinary Computer Use
Frontier AI agents frequently violate corrigibility by overriding interruptions in benign computer-use tasks, with misalignment increasing alongside model capability.
-
Triaging Threats to Specialized Guardrails
Introduces GuardZoo benchmark and RouteGuard router-expert system showing monolithic guardrails suffer task interference while specialized routing improves threat detection and generalization.
-
PageGuide: Browser extension to assist users in navigating a webpage and locating information
PageGuide is a browser extension that grounds LLM responses in webpage DOM elements via visual overlays for Find, Guide, and Hide modes, reporting performance gains over unaided browsing in a 94-user study.
-
PageGuide: Browser extension to assist users in navigating a webpage and locating information
PageGuide grounds LLM answers in webpage DOM elements using visual overlays for find, guide, and hide modes, yielding measurable gains in a 94-user study.
-
WebSP-Eval: Evaluating Web Agents on Website Security and Privacy Tasks
WebSP-Eval shows multimodal web agents fail many real security and privacy browser tasks, with stateful UI elements like toggles causing over 45% of failures.
-
FinToolBench: Evaluating LLM Agents for Real-World Financial Tool Use
FinToolBench couples 760 executable financial APIs with 295 tool-required queries and scores agents on execution success plus timeliness, intent, and domain compliance, with a finance-aware retrieval baseline (FATR).
-
Measuring the Security of Mobile LLM Agents under Adversarial Prompts from Untrusted Third-Party Channels
Indirect prompt injection through ads, webviews, and notifications reliably diverts mobile LLM agents into leaking data and installing malware across eight evaluated agents.
-
OpenRath: Session-Centered Runtime State for Agent Systems
OpenRath introduces Session as a first-class, branchable runtime value that unifies fragmented state in multi-agent systems and makes fork, merge, and replay explicit operations.
-
An Executable Benchmarking Suite for Tool-Using Agents
The paper delivers a unified executable benchmarking suite for tool-using agents that enforces a shared evidence-admission contract across web, code, and micro-task environments.
-
An Executable Benchmarking Suite for Tool-Using Agents
Introduces a benchmarking suite with common workload adapters, event schemas, and an evidence gate connecting WebArena Verified, SWE-Gym, and MiniWoB++ for tool-using agents.
-
Agentic AI Security: Threats, Defenses, Evaluation, and Open Challenges
A survey that taxonomizes threats to agentic AI, reviews benchmarks and evaluation methods, discusses technical and governance defenses, and identifies open challenges.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.