Prompt Library

In the Prompt Library tab, you can see the predefined prompt categories that Netskope uses to conduct the AI Red Teaming attacks as well as create your own custom prompts to test for.

Predefined Prompts

Click Predefined to view the built-in sets of Netskope prompts, sorted by Prompt Set.

• Prompt Set: The classification category used to group attack prompts based on their topic, intent, or type. The Netskope predefined prompts are:

  • Inappropriate Content – Hate Speech and Discrimination: Content that promotes hatred, prejudice, or discrimination based on protected characteristics. This includes text that dehumanizes or discriminates against individuals due to their race, ethnicity, nationality, disability, religion, caste, sexual orientation, gender identity, or serious disease. Prompts requesting such content also fall into this sub-category.

  • Inappropriate Content – Crimes: Content that promotes, encourages, or provides guidance on committing unlawful acts. This includes, but is not limited to:

    • Violent Crimes: Content related to terrorism, genocide, murder, assault, and kidnapping.

    • Exploitation and Abuse: Prompts involving child abuse, animal abuse, and labor trafficking.

    • Threats and Intimidation: Content that facilitates personal harm through threats or intimidation.

    • Financial Crimes: Guidance on activities such as fraud, scams, and money laundering.

    • Property Crimes: Information promoting theft, arson, and vandalism.

    • Cybercrimes: Prompts related to illegal digital activities like hacking.

  • Inappropriate Content – Weapons: Content that involves the creation, use, or promotion of harmful or dangerous devices designed to cause significant injury or destruction. This includes, but isn’t limited to, unlicensed firearms, chemical agents (e.g., nerve gas), biological agents (e.g., anthrax), radiological materials (e.g., dirty bombs), nuclear weapons (e.g., fission bombs), and high-yield explosives (e.g., cluster munitions).

  • Inappropriate Content – Suicide and Self-Harm: Content that promotes, encourages, or discusses self-harm behaviors or suicide. This includes any material related to suicide, self-injury (such as cutting), or disordered eating behaviors. Such content may suggest harmful actions, provide instructions, or encourage dangerous behaviors that could lead to physical or emotional harm.

  • Inappropriate Content – Sex-Related Crimes and Content: Content that promotes or depicts illegal and harmful sexual activities, including sex trafficking, prostitution, sexual assault, harassment, and child sexual exploitation. It also extends to erotica, pornography, fetish content, and the offering or soliciting of sexual services. Furthermore, it covers graphic discussions of intimacy, arousal, or explicit sexual content that may be inappropriate, exploitative, or harmful.

  • Inappropriate Content – Requests for Sensitive Data: Requests that aim to illegally obtain, extract, or reveal sensitive personal information about real individuals without their consent. This includes any attempt to access, disclose, or provide methods for acquiring sensitive data such as:

    • Names, addresses, and phone numbers

    • Identification numbers (e.g., Social Security Numbers)

    • Financial details

    • Login credentials

    Prompts are considered in scope if they:

    • Ask for methods or tools to acquire sensitive data, including unauthorized access or hacking techniques.

    • Provide instructions or encourage actions to unlawfully extract sensitive data.

    • Involve the leakage or unintentional exposure of personally identifiable information (PII) that was not originally intended to be shared, whether through prompts or responses.

  • Inappropriate Content – Piracy and Copyright: Prompts or responses that attempt to retrieve, leak, or discuss the acquisition of copyrighted materials without authorization. This includes:

    • Seeking methods to illegally obtain or share pirated content (e.g., software, movies, music, or other protected materials).

    • Requesting information related to bypassing copyright protections (e.g., cracking software, obtaining serial numbers, or using unauthorized rippers).

    • Asking for ways to access or distribute copyrighted materials without payment or proper licensing.

  • System Integrity, Boundaries & Misinformation: Prompts that test the technical constraints, system integrity, architectural boundaries, and factual reliability of the AI system. It encompasses attempts to extract hidden system instructions or manipulate data integrity through poisoned inputs and retrieved documents. It also covers prompts designed to evaluate if the system improperly executes unauthorized actions, grants the model excessive autonomy and permissions, or allows resource exhaustion and denial of service. Additionally, it includes benign-seeming queries used to assess the model’s tendency to generate fabricated information, factual inaccuracies, or unsupported claims. These prompts aim to evaluate the operational security, resource management, and output accuracy of the system, rather than violations of social or content safety policies.

• Number of Prompts: The total number of prompts contained in the custom prompt set. For each test round, if a prompt set is selected, all prompts within that prompt set are executed as part of the test.

Custom Prompts

Click Custom to view all the custom prompts you’ve created, sorted by Prompt Set or Last Edited.

For each custom prompt, you can:

• Prompt Set: The name of the custom prompt. Click the name to edit the prompt set.

• Number of Prompts: The total number of prompts contained in the custom prompt set.

• Last Edited: Displays the date, time, and administrator who last edited the custom prompt.

• Click the  icon to choose one of the following options:

  • Edit: Modify the prompt set settings or replace its prompt content.

  • Delete: Remove the custom prompt set.