Code execution via unsandboxed model chat templates
Published May 10, 2024 · Updated Aug 2, 2024
Server-side template injection in abetlen llama-cpp-python 0.2.30 through 0.2.71 allows attackers to execute code via a malicious GGUF model. Jinja2ChatFormatter compiles tokenizer.chat_template metadata with an unsandboxed Jinja2 Environment and renders it while constructing a chat prompt. Exploitation requires an application to load the attacker's model and invoke chat completion, after which commands run with the application's privileges.
Summary
What happened
Server-side template injection in abetlen llama-cpp-python 0.2.30 through 0.2.71 allows attackers to execute code via a malicious GGUF model. Jinja2ChatFormatter compiles tokenizer.chat_template metadata with an unsandboxed Jinja2 Environment and renders it while constructing a chat prompt. Exploitation requires an application to load the attacker's model and invoke chat completion, after which commands run with the application's privileges.
The record
- CVE
- CVE-2024-34359
- Published
- May 10, 2024
- Updated
- Aug 2, 2024
- Vendor
- abetlen
- Product
- llama-cpp-python
- Classifications
- CWE-76, T1204.002
- Attack vector
- network
- Privileges
- unauthenticated
Timeline
How it unfolded
- May 10, 2024CVE publishedPublication date reported by the CVE source.
- Aug 2, 2024Record updatedLatest update available in the CVE record.
Exploitability
Present is not the same as exploitable
Compare your product and version with the public record. A matching version still requires validation against your environment.
Is a vulnerable build present?
Compare these published version ranges with your installed build and any vendor patches.
- Affected versionversion=>= 0.2.30, <= 0.2.71
- Affected versionversion=0.2.30 <=0.2.71
What conditions does exploitation require?
What is affected?
Published CVSS scores
CVSS describes severity. EPSS estimates exploitation probability.
Attacks
What attackers are doing with it
Daily unique IPs observed by Shadowserver honeypots for known exploited vulnerabilities (KEVs). Missing observations do not establish an absence of attacks.
Weakness, pattern, technique
Public exploit references
- Malicious GGUF chat-template proof of conceptfunctional · demonstrated
Labels summarize the accepted research assessment. They do not indicate a test against your environment.
Technologies
Your stack
See the directory against your own environment.
Your stack
Check the software in your environment
Book a demo to see how Hinoki identifies affected software and validates exploitability in your environment.
Book a demo