Executive brief
aiwaves-cn agents is an open-source framework for building autonomous AI language agents. A vulnerability in how the system processes incoming messages allows an unauthenticated attacker to crash the backend service by sending an excessively large text payload. This results in a denial-of-service (DoS) condition, making the AI agent unavailable to legitimate users.
Technical details
An uncontrolled resource consumption vulnerability exists in the cheshire_cat_core component of aiwaves-cn agents. The 'recall_relevant_memories_to_working_memory' function in 'core/cat/looking_glass/stray_cat.py' processes user-supplied text through the 'tiktoken' library to calculate token counts without enforcing length limits. An unauthenticated remote attacker can send a specially crafted, oversized JSON payload (e.g., 5 million characters) to the '/message' endpoint. This triggers a stack overflow (RuntimeError) in the underlying Rust-based tiktoken library, causing the Python FastAPI worker process to crash. As of the advisory date, the project follows a rolling release model and has not yet released a formal patch for this issue.
Affected products
- aiwaves-cn agents <= 0.0.52
- aiwaves-cn cheshire_cat_core <= 0.0.52
Timeline
- 2026-03-23: other: Issue reported to the project maintainers
- 2026-05-11: advisory: GitHub Advisory published
- 2026-05-11: disclosed: CVE-2026-8319 assigned