Security audit
nanogpt-training
Security checks for vulnerabilities and agentic risk
Overview
This is a text-only guide for training small GPT-style models and does not show hidden, destructive, or credential-seeking behavior.
Before installing, consider the practical cost of using the examples: they can download hundreds of megabytes or more of public training data and may use paid cloud GPU resources. Only provide HuggingFace, Modal, or cloud credentials if you intentionally choose private resources or remote execution.
Vulnerability Patterns
- Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
- Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
- Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
- Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Static analysis
No suspicious patterns detected.