Call for Papers
Call for abstracts
The Workshop on Formal Methods for Agentic AI (FMAAI) aims to bridge traditional formal methods and the emerging field of agentic software systems. We seek to explore how formal models and specifications (temporal logics, state machines, invariants) can provide the guardrails that let AI agents operate autonomously without violating critical properties such as safety, security, and privacy.
Topics of interest include, but are not limited to:
- Auto-formalization: using LLMs and formal tools to extract specifications and models from existing codebases
- Neurosymbolic observability: combining machine learning with formal verification to monitor and control runtime behavior
- Agent guardrails: policy-driven constraints ensuring agent actions satisfy safety invariants
- Deterministic simulation for agents, e.g., for training reinforcement learning agents on complex failure scenarios
- State machine and workflow inference from noisy historical trajectories and distributed workflows
- Runtime verification and logic-grounded planning: verifying agent actions and automated repair with formal explanations
- Agent operating systems and proof-carrying code: verification-aware instrumentation for formal agentic grounding
Submission details
- Extended abstracts of up to 2 pages.
- Submissions are reviewed by the program committee.
- Accepted abstracts are invited for a 10-minute lightning talk and a poster presentation.
- Accepted extended abstracts and presentation slides will be published open-access on the workshop website.
- Submission portal: To be announced.