Back to the MIT repository
2. Privacy & Security1 - Pre-deployment

Programming Language

Most LLMs are developed using the Python language, whereas the vulnerabilities of Python interpreters pose threats to the developed models

Source: MIT AI Risk Repositorymit20

ENTITY

3 - Other

INTENT

2 - Unintentional

TIMING

1 - Pre-deployment

Risk ID

mit20

Domain lineage

2. Privacy & Security

186 mapped risks

2.2 > AI system security vulnerabilities and attacks

Mitigation strategy

1. Adopt the principle of least privilege in code execution by utilizing internal Python APIs for functionality instead of invoking Operating System (OS) commands, and employ secure execution environments (sandboxing) for any dynamically generated or untrusted code to contain potential threats. 2. Implement a robust software supply chain security protocol, which includes continuous vulnerability scanning of all project dependencies, maintaining the minimal viable runtime environment, and applying security patches promptly to mitigate risks associated with third-party packages. 3. Enforce strict input validation and sanitization mechanisms on all user-supplied data to prevent injection attacks; specifically, utilize parameterized statements for database interactions and array-based arguments for subprocess execution to logically separate data from code.