Search papers, labs, and topics across Lattice.
AutoRestTest addresses the challenges of black-box REST API testing by integrating a Semantic Property Dependency Graph with multi-agent reinforcement learning and large language models to systematically explore extensive input spaces. This approach proved to be highly effective, as evidenced by its top ranking in fault detection, efficiency, and effectiveness during the SBFT 2026 REST League competition, where it evaluated 11 APIs with a total of 317 operations. The system averaged 67.09 unique server errors and successfully processed 17.27 operations per API within a one-hour testing window, showcasing its robust capabilities in API testing.
AutoRestTest outperformed all competitors in the SBFT 2026 Tool Competition, revealing significant advancements in automated REST API testing.
Large input spaces and complex inter-operation dependencies make black-box REST API testing challenging. AutoRestTest combines a Semantic Property Dependency Graph, multi-agent reinforcement learning, and large language models to intelligently explore large API input spaces. In the SBFT 2026 REST League, AutoRestTest ranked first in all three evaluation categories -- fault detection, overall efficiency, and overall effectiveness -- on 11 APIs (317 operations, approximately 29 per API), averaging 67.09 unique server errors and 17.27 successfully processed operations per API under a one-hour testing budget.
Amazon Science