Raw vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:HSummary
CVE-2026-40088 is a critical-severity OS Command Injection (CWE-78) vulnerability in Praison Praisonai. Its CVSS base score is 9.6 (Critical).
Operationally, exploitation aligns with the MITRE ATT&CK technique Command and Scripting Interpreter (T1059); ranked at the 35th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog; a public proof-of-concept is referenced.
This vulnerability is AI-related — categorised as AI Agent Protocols and Integrations; in the LLM/Generative AI Risks risk domain.
The strongest mitigations our analysis identified map to SA-11 (Developer Testing and Evaluation) and SI-10 (Information Input Validation) — see the control section below for these in your framework.
Deeper analysis AI-assisted summary
Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.
CVE-2026-40088 is a command injection vulnerability (CWE-78) in PraisonAI, a multi-agent teams system, affecting versions prior to 4.5.121. The execute_command function and workflow shell execution components are exposed to user-controlled input via agent workflows, YAML definitions, and LLM-generated tool calls. This exposure allows attackers to inject arbitrary shell commands through shell metacharacters.
Unauthenticated remote attackers can exploit the vulnerability over the network with low complexity and no privileges required, though user interaction is needed. By crafting malicious inputs such as workflows or tool calls, attackers can trick users into triggering execution, achieving arbitrary shell command injection on the host system. This results in high impacts to confidentiality, integrity, and availability, with a changed scope, yielding a CVSS v3.1 base score of 9.6.
The vulnerability is addressed in PraisonAI version 4.5.121. The GitHub security advisory (GHSA-2763-cj5r-c79m) and release notes for v4.5.121 detail the fix and recommend upgrading immediately.
This vulnerability has relevance to AI/ML deployments, as it involves LLM-generated tool calls in a multi-agent framework. No public information on real-world exploitation is available.
OWASP Top 10 for Web (2025)
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2026-21064
Vulnerability Data
PraisonAI is a multi-agent teams system. Prior to 4.5.121, the execute_command function and workflow shell execution are exposed to user-controlled input via agent workflows, YAML definitions, and LLM-generated tool calls, allowing attackers to inject arbitrary shell commands through shell metacharacters.…
more
This vulnerability is fixed in 4.5.121.
- CWE(s)
AI Security AnalysisAI
- AI Category
- AI Agent Protocols and Integrations
- Risk Domain
- LLM/Generative AI Risks
- OWASP Top 10 for LLMs 2025
- None mapped
- AI-specific weaknesses CR
- CWE-1426 — LLM-generated tool calls reach OS command sink without validation
Mapped by Cyber Resilience · not in NVD. Poisoning and extraction cases are routed to MITRE ATLAS instead of a synthetic CWE.- Classification Reason
- Matched keywords: llm
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
—
V1.2.5V1.2.8V15.2.5
Mitigating Controls (NIST 800-53 r5) AI
Developer testing and evaluation can discover missing or incorrect command sanitization during development.
Input validation directly neutralizes or rejects special characters that would otherwise alter OS command structure.
Least privilege reduces the permissions available to any process that could be subverted by injected commands.
Least functionality restricts available OS commands and interpreters, limiting the blast radius of injection.
Secure engineering principles require proper neutralization of untrusted input before command construction.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
PR.PS-06's SDLC practices directly require secure coding and input handling that blocks command-injection defects, yet the single broad outcome leaves many specific neutralization vectors and verification gaps unaddressed.
Routine patching/maintenance can remediate known command-injection CVEs in dependencies (partial forward) but does nothing to stop developers from introducing improper neutralization in custom code (none reverse).
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
Security testing and code review target insecure use of operating-system command interfaces, catching command-injection flaws introduced during development.