Related papers: On the Reliability of Computer Use Agents
The benchmarks used to evaluate AI agents in security-critical roles suffer from crucial weaknesses. Building on recent empirical evidence, we characterize three core challenges that undermine security evaluations: benchmark…
Agentic systems are evaluated on benchmarks where agents interact with environments to solve tasks. Most papers report a pass@1 score computed from a single run per task, assuming this gives a reliable performance estimate. We test this…
AI agents are an exciting new research direction, and agent development is driven by benchmarks. Our analysis of current agent benchmarks and evaluation practices reveals several shortcomings that hinder their usefulness in real-world…
This paper studies algorithmic decision-making under human's strategic behavior, where a decision maker uses an algorithm to make decisions about human agents, and the latter with information about the algorithm may exert effort…
Unlike traditional automation tools or static LLM-based systems, agents combine decision-making and tool utilization to accomplish complex tasks, showing great potential in software engineering. However, existing studies largely focus on…
Tool-using agents often incur substantial computational cost due to long reasoning chains and iterative tool usage. In practical scenarios, many tasks become infeasible under constrained tool environments, where the capabilities required…
AI agents are increasingly deployed in multi-task settings, where the task to perform is specified at test time, and the agent must generalize to unseen tasks. A major concern in such settings is safety: often, an agent must not only…
Modern AI agents increasingly combine conversational interaction with autonomous task execution, such as coding and web research, raising a natural question: What happens when an agent engaged in long-horizon tasks is exposed to user…
Large language model-based multi-agent systems have shown great abilities across various tasks due to the collaboration of expert agents, each focusing on a specific domain. However, the impact of clumsy or even malicious agents--those who…
Classic evaluation methods of believable agents are time-consuming because they involve many human to judge agents. They are well suited to validate work on new believable behaviours models. However, during the implementation, numerous…
Recent advances in browser-based LLM agents have shown promise for automating tasks ranging from simple form filling to hotel booking or online shopping. Current benchmarks measure agent performance in controlled environments, such as…
When a phone-use agent avoids harm, does that show safety, or simply inability to act? Existing evaluations often cannot tell. A harmful outcome may be avoided because the agent recognized the risk and chose the safe action, or because it…
The problem `human and work' in a model working group is investigated by means of cellular automata technique. Attitude of members of a group towards work is measured by an indicator of loyalty to the group (the number of agents who carry…
Research on multi-agent planning has been popular in recent years. While previous research has been motivated by the understanding that, through cooperation, multi-agent systems can achieve tasks that are unachievable by single-agent…
Browser agents enable autonomous web interaction but face critical reliability and security challenges in production. This paper presents findings from building and operating a production browser agent. The analysis examines where current…
Interactive agent benchmarks map an agent run to a binary outcome through outcome checks. When these checks rely on surface level signals or fail to capture the agent's actual action path, they cannot reliably determine whether the run…
In order for agents trained by deep reinforcement learning to work alongside humans in realistic settings, we will need to ensure that the agents are \emph{robust}. Since the real world is very diverse, and human behavior often changes in…
We take the position that agent security must be approached as a systems problem: the AI model powering the agent must be treated as an untrusted component, and security invariants must be enforced at the system level. Through this lens,…
AI Agents can perform complex operations at great speed, but just like all the humans we have ever hired, their intelligence remains fallible. Miscommunications aren't noticed, systemic biases have no counter-action, and inner monologues…
Autonomous agent systems powered by Large Language Models (LLMs) have demonstrated promising capabilities in automating complex tasks. However, current evaluations largely rely on success rates without systematically analyzing the…