Skip to content

Deduplicate unsafe-chain scan results #9

Open
@rbitr

Description

@rbitr

Currently duplicate unsafe chains will often be detected, example below. The tool should only report the longest such chain

Location: ../demo_scans/GenAI-Showcase/apps/local-rag-pdf/rag_module.py:181
Message: Untrusted input 'query' flows to LLM API call without proper sanitization
Tags: security, sanitization, prompt-engineering
Suggestion: Implement input validation or sanitization before passing untrusted input to LLM. Consider using an allow-list approach.
Context:
  source: query
  sink: llm_call_181
  path: query -> retrieved_docs -> formatted_input -> llm_call_181

Issue #11: chain-unsafe-input (high)
Location: ../demo_scans/GenAI-Showcase/apps/local-rag-pdf/rag_module.py:181
Message: Untrusted input 'question' flows to LLM API call without proper sanitization
Tags: security, sanitization, prompt-engineering
Suggestion: Implement input validation or sanitization before passing untrusted input to LLM. Consider using an allow-list approach.
Context:
  source: question
  sink: llm_call_181
  path: question -> retrieved_docs -> formatted_input -> llm_call_181

Issue #12: chain-unsafe-input (high)
Location: ../demo_scans/GenAI-Showcase/apps/local-rag-pdf/app.py:61
Message: Untrusted input 'user_input' flows to LLM API call without proper sanitization
Tags: security, sanitization, prompt-engineering
Suggestion: Implement input validation or sanitization before passing untrusted input to LLM. Consider using an allow-list approach.
Context:
  source: user_input
  sink: llm_call_61
  path: user_input -> llm_call_61

Issue #13: chain-unsafe-input (high)
Location: ../demo_scans/GenAI-Showcase/apps/local-rag-pdf/app.py:61
Message: Untrusted input 'user_input' flows directly to LLM API call without proper sanitization
Tags: security, sanitization, prompt-engineering
Suggestion: Implement input validation or sanitization before passing untrusted input to LLM. Consider using an allow-list approach.
Context:
  source: user_input
  sink: llm_call_61
  path: user_input -> llm_call_61

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions