Neurips 2024 Workshop
Towards Safe & Trustworthy Agents

This workshop aims to clarify key questions on the safety of agentic AI systems and foster a community of researchers working in this area.

To this end, we have prepared a diverse and comprehensive schedule of speakers and organizers, to bring together experts in the field. We are also making a call for papers below, with submissions also accepted through the 'submit' button above.

Please note that the order of speakers in the schedule below is subject to change depending on speaker schedule and availability. We will have a confirmed speaker order closer to the time.

‍

Call for papers deadline: September 14th, 2024

The schedule above is tentative and the order of speakers could change.

Foundation models are increasingly being augmented with new modalities and access to a variety of tools and software [9, 5, 3, 11]. Systems that can take action in a more autonomous manner have been created by assembling agent architectures or scaffolds that include basic forms of planning and memory, such as ReAct [12], RAISE [7], Reflexion [10], and AutoGPT + P [2], or multi-agent architectures such as DyLaN [8] and AgentVerse [4]. As these systems are made more agentic, this could unlock a wider range of beneficial use-cases, but also introduces new challenges in ensuring that such systems are trustworthy [6]. Interactions between different autonomous systems create a further set of issues around multi-agent safety [1]. The scope and complexity of potential impacts from agentic systems means that there is a need for proactive approaches to identifying and managing their risks. Our workshop will surface and operationalize these questions into concrete research agendas.

‍

This workshop aims to clarify key questions on the trustworthiness of agentic AI systems and foster a community of researchers working in this area. We welcome papers on topics including, but not limited to, the following:

‍
• Research into safe reasoning and memory. We are interested in work that makes LLM agent
reasoning or memory trustworthy, e.g., by preventing hallucinations or mitigating bias.
• Research into adversarial attacks, security and privacy for agents. As LLM agents interact
with more data modalities and a wider variety of input/output channels, we are interested in work
that studies or defends against possible threats and privacy leaks.
• Research into controlling agents. We are interested in novel control methods which specify goals,
constraints, and eliminate unintended consequences in LLM agents.
• Research into agent evaluation and accountability. We are interested in evaluation for LLM
agents (e.g., automated red-teaming) and interpretability + attributability of LLM agent actions.
• Research into environmental and societal impacts of agents. We are interested in research that
examines the environmental cost, fairness, social influence, and economic impacts of LLM agents.
• Research into multi-agent safety and security. We are interested in research that analyzes novel
phenomena with multiple agents: emergent functionality at a group level, collusion between agents, correlated failures, etc.

‍

• Submission site: Submissions should be made on OpenReview.
• Submissions are non-archival: we receive submissions that are also undergoing peer review elsewhere at the time of submission, but we will not accept submissions that have already been previously published or accepted for publication at peer-reviewed conferences or journals. Submission is permitted for papers presented or to be presented at other non-archival venues (e.g. other workshops). No formal workshop proceedings will be published.
• Social Impact Statement: authors are required to include a "Social Impact Statement" that highlights "potential broader impact of their work, including its ethical aspects and future societal consequences".
• Submission Length and Format: Submissions should be anonymised papers up to 5 pages (appendices can be added to the main PDF); excluding references and Social Impacts Statement. You must format your submission using the NeurIPS 2024 style file.
• Paper Review: All reviews are double-blinded, with at least two reviewers assigned to each paper.
• Camera Ready Instructions: The camera ready version is composed of a main body, which can be up to 6 pages long, followed by unlimited pages for a Social Impact Statement, references and an appendix, all in a single file. The camera-ready versions of all accepted submissions should be uploaded by the authors to the OpenReview page for corresponding submissions. The camera ready version will be publicly available to everyone on the Camera-Ready Deadline displayed below.