/headshot.jpg
cloud infrastructure • ramblings • software

Read the Fine Print: AI Data Usage

AI tools offer significant productivity gains. The cost isn’t just your monthly subscription fee.

What You’re Actually Agreeing To

Most AI service terms allow companies to use your prompts and data for model training. Your code snippets, business logic, and sensitive information become part of their training datasets.

Common Terms of Service Clauses:

  • Prompts and responses used for “service improvement”
  • Data retention periods measured in years, not months
  • Broad permissions for “research and development”
  • Right to share data with third-party partners
  • Minimal user control over data deletion

Data That Gets Collected

Beyond Your Prompts:

Don't Trust AI to Change Your Systems

AI coding assistants are powerful tools for understanding and analyzing codebases. The challenge is balancing their analytical capabilities with system safety.

The Risk of Automatic Write Access

AI models hallucinate, make assumptions about business logic, and lack context about system dependencies. A single automated change can break production, corrupt data, or introduce security vulnerabilities.

Common AI Mistakes:

  • Deleting files it thinks are unused
  • Moving configuration files to “better” locations
  • Modifying permissions without understanding security implications
  • Creating directories that conflict with existing deployment scripts
  • Copying files with incorrect ownership or permissions

The Solution: Read-Only Access

Give AI comprehensive read access to gather information quickly. Require human approval for all modifications.