Security measures (Model Code Safety)

[schuelet]

Current State:

• The models provided as FSKX file contain the executable model script. If this code is unsupervised it could be a security risk for the executing platform.

  Goal:

• The general code safety should be strengthened by additional measures. The FSK-ML should be able to prevent unwanted behavior of the code, like loading hidden resources or manipulate the executing system.

Concept:

Implementation:

Challenges:

• Improving code safety for 'unsupervised' models.

  Resources:

• Available: 2 developer
• KNIME: 50 days
• Testing: 5 days

Total Cost:

55 days

[schuelet]

Strategies to handle script code:

• whitelist of permitted commands
  • too specific, most models won't run
• blacklist of prohibited commands
  • can be used to lock down the most commonly used attempts at introducing harmful code
  • non-exhaustive, prone to be worked around
• manual curation (code review & test) by trusted community members
  • time consumptive
  • testing large scripts requires save environments for curators