Hex Encoding Strategy

The Hex Encoding strategy tests an AI system's ability to resist ASCII hex-encoded inputs that might bypass security controls. Hex encoding (using characters 0-9, A-F) exploits a potential gap in safety mechanisms by representing text as a sequence of hexadecimal bytes.

Why It Works

• Models may learn hex encoding patterns from training data
• Safety training rarely includes encoded content, making hex inputs out-of-distribution
• More capable models may be able to decode hex values, potentially bypassing safety filters
• Safety training may not scale proportionally with model capabilities for encoded inputs

Implementation

Use it in your promptfooconfig.yaml:

strategies:
  - hex

Example Scenarios

The strategy helps identify:

• Security filter bypasses through hex-encoded content
• Model's ability to handle obfuscated malicious inputs
• Potential jailbreaking vulnerabilities through encoding

Related Concepts

• Base64 Encoding - Similar encoding strategy
• ROT13 Encoding - Alternative encoding strategy
• Leetspeak - Text obfuscation technique

For a comprehensive overview of LLM vulnerabilities and red teaming strategies, visit our Types of LLM Vulnerabilities page.