Research demo
LLMSmith: 20 RCE vulnerabilities across 11 LLM agent frameworks, exploited via prompt injection
AIR-2023-0021 AIR record · 2023-09-llmsmith-framework-rce
From the supporting source
Our research has uncovered a total of 20 vulnerabilities in 11 LLM-integrated frameworks, comprising 19 RCE vulnerabilities and 1 arbitrary file read/write vulnerability.arxiv.org
The record
Disclosed
2023-09-06
Vendor
multiple open-source projects
Product
LangChain, LlamaIndex, Pandas-ai, Langflow, Auto-GPT, Griptape, Lagent, MetaGPT, vanna, langroid, Pandas-llm
Risk domain
Access control and permissions › Privilege escalation
System type
Generative model
Who acted
Agent acted — injected prompts drove code execution inside the frameworks
Attack surface
Agent framework
Autonomy level
Semi-autonomous
Initial vector
Direct prompt injection user input
Impact type
Code execution, Arbitrary file read write
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI03 – Identity and Privilege Abuse
- MITRE ATLAS
- AML.T0055 – Unsecured Credentials
- NIST AI RMF
- GOVERN 6.1
- EU AI Act
- Article 13
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool misuse
- Mitigation
- Apply least-privilege principle; Regular policy audits.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
51 internet-exposed apps identified; 17 successfully exploited
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
17 of 20 confirmed by maintainers; 11 CVEs assigned
Taxonomy & references
Curator notes
Tong Liu et al. ATLAS dates this to the v4 revision (2025-02-27); original submission 2023-09-06, which is used as date_disclosed.
Supporting sources
Cite as
Enkrypt AI. "LLMSmith: 20 RCE vulnerabilities across 11 LLM agent frameworks, exploited via prompt injection". AIR, AIR-2023-0021. https://agentincident.vercel.app/i/AIR-2023-0021/ (accessed 2026-10-19).