System Instructions Extraction

Type: technique

Description: The adversary extracts system instructions provided by the AI system builder to learn about the system's capabilities and circumvent its guardrails.

Version: 0.1.0

Created At: 2024-10-03 22:24:49 +0300

Last Modified At: 2024-10-03 22:24:49 +0300

External References

• Extracting Microsoft Copilot's System Instructions, Zenity Labs

Related Objects

• --> Discovery (tactic)
• <-- System Instruction Keywords (technique): Extracting the system instructions is a pre-requisite to applying this technique
• <-- GitHub Copilot Chat: From Prompt Injection to Data Exfiltration (procedure): Extracting system instructions reveal that GitHub Copilot is powered by GPT4.

Related Frameworks