# CVE-2026-108768: zhayujie CowAgent resource exhaustion in tool-call argument parsing

Source: https://junglewise.ai/threats/cve-2026-108768-a-vulnerability-was-identified-in-zhayujie-cowagent-up-to-2-2-0  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-11. Updated: 2026-10-16  
License: CC BY 4.0 with a link back

CVE-2026-108768 · Severity: medium · CVSS 4.3 · Published 2026-10-11

CowAgent is an AI agent framework that processes function calls from language models. A denial-of-service vulnerability allows a malicious or compromised model provider to cause unbounded memory and CPU consumption by returning oversized tool-call arguments, making the service unavailable or unresponsive to legitimate requests.

Technologies: [Zhayujie CowAgent](https://junglewise.ai/threats/technologies/cowagent). Vendors: [Zhayujie](https://junglewise.ai/threats/vendors/zhayujie).

## Quick answers

**Is CVE-2026-108768 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-108768 being exploited in the wild.

**How severe is CVE-2026-108768?** Medium severity, with a CVSS score of 4.3, published 11 October 2026.

**Which products and versions are affected?** Zhayujie CowAgent up to 2.2.0.

**Is there a fix for CVE-2026-108768?** No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-108768 attacked?** It is attacked over the network.

**Am I affected by CVE-2026-108768?** Check whether you run Zhayujie CowAgent at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## Technical details

The vulnerability is an uncontrolled resource consumption (CWE-400) in the streaming tool-call argument handler. When a model streams function arguments in chunks, agent/protocol/agent_stream.py accumulates the entire argument string without size limits before calling json.loads() and json_repair.repair_json(). A remote attacker controlling or compromising a model provider can inject arbitrarily large tool arguments over the network, causing unbounded memory allocation and CPU-intensive parsing operations.

## Affected products

- zhayujie CowAgent up to 2.2.0

## Timeline

- 2026-10-11: disclosed

## References

- [https://github.com/lche511/cve/tree/main/CowAgent_Tool_Call_Arguments_JSON_Repair_DoS_Report](https://github.com/lche511/cve/tree/main/CowAgent_Tool_Call_Arguments_JSON_Repair_DoS_Report)
- [https://vuldb.com/cve/CVE-2026-108768](https://vuldb.com/cve/CVE-2026-108768)

## Related threats

- [zhayujie CowAgent denial of service in Markdown rendering](https://junglewise.ai/threats/cve-2026-108769-a-security-flaw-has-been-discovered-in-zhayujie-cowagent-up-to-2)
- [zhayujie CowAgent uncontrolled memory allocation in media download](https://junglewise.ai/threats/cve-2026-108683-a-security-vulnerability-has-been-detected-in-zhayujie-cowagent)
- [zhayujie CowAgent denial of service in /upload endpoint](https://junglewise.ai/threats/cve-2026-108682-a-weakness-has-been-identified-in-zhayujie-cowagent-up-to-2-1-6)
- [zhayujie CowAgent denial of service in Web Console](https://junglewise.ai/threats/cve-2026-108681-a-security-flaw-has-been-discovered-in-zhayujie-cowagent-up-to-2)
- [zhayujie CowAgent denial of service in Bash tool](https://junglewise.ai/threats/cve-2026-84427-zhayujie-cowagent-denial-of-service-in-bash-tool)