GHSA-J659-8XH6-5PQ5: Financial Guardrail Bypass in atomic-agents-stack via Parallel Execution of Unlisted Models
Vulnerability ID: GHSA-J659-8XH6-5PQ5
CVSS Score: 8.7
Published: 2026-08-17
A high-severity vulnerability in the atomic-agents-stack framework allows complete bypass of cost-cap guardrails during parallel model execution when utilizing unlisted, local, or self-hosted models.
TL;DR
Unrecognized LLMs evaluate to a $0.00 reservation cost, completely bypassing batch cost guardrails and allowing infinite execution spend.
Technical Details
- CWE ID: CWE-770
- Attack Vector: Network
- CVSS v4 Score: 8.7
- Exploit Status: Proof of Concept Available
- CISA KEV Status: Not Listed
- Affected Components: atomic_agents/agent.py
Affected Systems
- atomic-agents-stack
-
atomic-agents-stack: <= 1.0.0 (Fixed in:
1.1.0)
Mitigation Strategies
- Upgrade atomic-agents-stack to 1.1.0 or higher
- Manually inject custom pricing definitions into the _costs.PRICING registry at application initialization
Remediation Steps:
- Identify the current installed version of atomic-agents-stack using command line checks.
- Upgrade the package using standard package managers (pip install --upgrade atomic-agents-stack>=1.1.0).
- Verify installation of the patched version by executing import assertions.
- For systems unable to upgrade, configure a startup script to manually inject custom model metadata into the PRICING dictionary.
References
- GitHub Security Advisory GHSA-j659-8xh6-5pq5
- Repository Advisory Details
- Project GitHub Release Notes (v1.1.0)
- Source Code Repository
Read the full report for GHSA-J659-8XH6-5PQ5 on our website for more details including interactive diagrams and full exploit analysis.
Top comments (0)