Exporting Results
Prompt Attack provides a streamlined feature for exporting adversarial prompt results into an Excel file, ensuring structured tracking, analysis, and reporting. This functionality is vital for security teams, particularly Blue Teams, as it enables systematic evaluation of threats, refinement of defenses, and overall enhancement of LLM security postures.
Populating the Prompt List
Before exporting, ensure the Prompt Attack module is populated with adversarial prompts. This module houses all the test data required for export, either through automatically generated prompts or manually crafted entries. Each entry typically contains the following attributes:
Identifies the category of the adversarial prompt, such as Security or Safety.
Specifies the adversarial approach employed, such as Prompt Injection, Prompt Probing, or Text Completion attacks.
The specific adversarial input tested against the LLM.
Describes the intended response of the LLM when confronted with the adversarial input.
Exporting Results to Excel / JSON
Exporting adversarial prompt results to CSV or JSON is a straightforward process. Follow these steps to complete the export:
Navigate to the Prompt Attack Section and locate the export button
Click on the Export button. This action triggers the system to compile all currently displayed entries into a downloadable Excel / JSON file.
The system captures only the data currently displayed in the module. Ensure that all desired entries you want to export are available in the table before initiating the export.
Download and Save
Once the export process is initiated:
The system will generate an Excel file in .csv and / or JSON file in .json format containing all adversarial prompt entries from the module.
The file will be automatically downloaded to the user's local machine.
The exported file can then be accessed offline for analysis, record-keeping, or sharing with other stakeholders.
CSV (Excel)
JSON
Use Cases for Exported Results
Exported data serves a critical role in several cybersecurity and compliance efforts. Below are practical applications of the exported Excel files:
Exported results support Blue Teams by providing detailed insights into adversarial testing activities. Teams can use these insights to:
- Analyze patterns in successful attacks (Exploited) prompts to identify weaknesses in the LLM's defenses.
- Fine-tune security filters and implement more robust response mechanisms to mitigate specific attack techniques.
Exported data enables structured reporting, which is vital for communicating testing outcomes and mitigation strategies to stakeholders:
- Generate clear and concise reports demonstrating the effectiveness of implemented security measures.
- Share results with internal or external teams to foster collaboration and transparency.
The ability to track adversarial testing over time provides valuable insights into evolving threats and system adaptations:
- Analyze trends in success/failure rates to identify recurring vulnerabilities or improvements in the LLM's resilience.
- Evaluate how the LLM adapts to emerging adversarial techniques and adjust defenses accordingly.
Try It Out with Prompt Attack
Exporting adversarial prompt testing results is an essential feature of Prompt Attack, providing security teams with the tools they need to track, analyze, and report on adversarial risks systematically. This functionality is particularly critical for Blue Teams focused on improving LLM defenses, refining mitigation strategies, and ensuring compliance with security standards.
By leveraging exported results, organizations can enhance their ability to detect, mitigate, and prevent adversarial threats. This contributes to building robust and resilient AI systems capable of withstanding evolving attack vectors in 2025 and beyond.