physical-ai-people-attribute-search

Pass

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Orchestrates physical-AI workflows using shell scripts and platform CLI tools like osmo and kubectl for setup, submission, and monitoring.
  • [COMMAND_EXECUTION]: Dynamically constructs execution arguments and configuration files at runtime using Python and shell scripts to manage per-sample augmentation parameters.
  • [EXTERNAL_DOWNLOADS]: Performs network probes to the nvcr.io container registry for image verification and communicates with OpenAI-compatible model endpoints for inference tasks.
  • [PROMPT_INJECTION]: Ingests and processes user-provided image datasets through LLM and VLM models during the augmentation and labeling stages, representing an inherent vector for indirect prompt injection via data poisoning.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 2, 2026, 11:23 PM
Security Audit — agent-trust-hub — physical-ai-people-attribute-search