What are Targets ?

In the Fortify ecosystem, The ‘Target’ refers to the specific AI-powered application, or Large language models(LLM), being evaluated for vulnerabilities, security risks, and ethical concerns. This system includes all components that interact with users, process data, and generate responses.

The targets section contains a list of all the targets created by the user for red teaming. Each target is listed with several columns: the target name, the target domain, the creation date and time, the name of the creator, the last modification date and time, the name of the person who last modified it, and an action button.

Active Target Tab

In the "Active Target" tab, all the targets created and active in the system so far are listed, along with details such as the created by, industry, version, integration type and other relevant information.

• On the Right side of the page (Highlighted in Red), the user is provided with an option to “Search” any target by ‘Name', 'Industry' or ‘Created By’ detail.
• Users can select the search criteria from a drop-down menu on the left side of the search section, enter the target name or industry, and click the magnifying glass button to search as highlighted above.

Filter Target

An option to Filter by ‘Integration Type’ and ‘Date Range’ is also available.

• The filter section has two options: ‘Integration Type’, where users can select targets integration method of their choice from a drop-down menu, and Date Range, where users can select a date range to view targets created within that period.
• After selecting filters, users must click Apply for them to be implemented. They can reset the filters by clicking Reset, or cancel the action by clicking on the ‘Cross’ sign.

Update/Edit Target

An ability to take Action is provided on the right side via the ‘View Details', 'Run Session’ and ‘Delete Target’ icons on the selected target (Highlighted in Green).

If the user clicks on ‘View’ details, it gets directed to “Target Configuration” page where the user can:

• View the existing target details.
• Add/Edit Code of Conduct.
• Set up HTTP Integration, Edit/Update Integration
• Regenerate/Delete API Key.

1. Here, by clicking on 'View Details' icon we can see the existing target details.

The Extracted Attack objectives can be seen on the left along with the Target specific details mentioned on the right.

2. A button to Edit Code of Conduct is provided to edit the terms of target code of conduct.

3. Update/Edit HTTP Integration Settings for a Target (Highlighted in Red) as seen below:

Drilling down on the HTTP integration(which is our default integration method), the users can edit or add to the integration settings.

4. Similarly, from the API integration drill down, the user can generate and copy the API key for integration.

Archived Target Tab

In the "Archived Target" tab, you can find all the targets that have been removed from the "Active Target" section. When a target is deleted from the active list, it moves to the ‘Archived’ section.

Filter Target

• The functionalities of “Search” and “Filter” are the same as in 'Active Target' tab.
• You can Filter by ‘Integration Type’ and ‘Date Range’ and Search any target by ‘Name', 'Industry' or ‘Created By’ detail.

Add New Target

• Moving ahead, on the Top right hand side of the Target page, the “Add New Target” button is present. (Highlighted in Red)

• As the user clicks on the "Add New Target" section, the target configuration screen appears.

1. The first step is to enter a unique and identifiable ‘Target Name’. Then, the user must specify the knowledge domain or industry to which the target belongs along with the 'Supported Language' of their choice. (English/French)(Highlighted in Red)
2. A new feature to ADD a icon to the Target has been introduced. We can choose an icon for the target from the available options. (Highlighted in Red)
3. Fallback Responses are predefined replies that a Target system should return when it encounters specific conditions such as:
   • Inability to answer a query.
   • Detection of a potential security threat.
   • Input outside the Target’s intended scope.

Thus, it assists the users to set Default replies for queries the system can't handle, threats, or off-topic requests.

Users need to fill out the Fallback Response section as mentioned in the above image (Highlighted in Red). This is a mandatory field.

By configuring expected fallback responses during Target setup, Fortify can better evaluate whether the Target system is correctly enforcing its safety mechanisms. The Judge Agent uses these configured responses as part of its evaluation criteria to verify that the Target appropriately:

• Handles edge cases.
• Refuses unsafe or inappropriate requests.
• Manages out-of-scope queries gracefully.

4. The user needs to complete one of the Integration method of their choice(i.e. HTTP integration or the API integration). This step is mandatory for our users.

5. Once the Integration step is completed, the Target gets created successfully in the system. A system pop-up message is displayed confirming successful Target creation. Refer to the below image:

The user then proceeds to fill out the Code of Conduct.

Code of Conduct

• The user is required to enter the code of conduct. They can also access a best practices guide for assistance. The best practices guide offers our users means to write effective Code of conduct.
• Code of Conduct is a set of rules and guidelines that the target system should adhere to, determined by the owner(s) of the target system. This is given to the Judge Agent to classify target system responses as vulnerabilities or not.
• A series of questions are provided for the user to set-up the Code of conduct.

• The user needs to fill in the following fields to set up the Code of conduct:

1. Target Purpose and Scope: Define the primary purpose of the target.
2. Approved Topics: Specify the advanced topics the target is authorized to engage with.
3. Prohibited Topics and Behaviors: Indicate the topics and behaviors the target should avoid.
4. Tone of Conversation: Set the tone the target should use.

• Once these details are filled in, the user can click "Save" to proceed or "Reset" to clear the entered information. Additionally, an option to “Skip” has been added for the user, to completely skip this step.

A prompt message will appear, briefing the user of the implications of skipping this step.

• Now, the created ‘Target’ can be seen in the Target Tab(Active). The User clicks on the ‘Target’ to ‘View Details’ of the selected Target as seen in the image below(Highlighted in Red)

• A new Target Details tab opens, showing all the target details previously filled by the user. An option to 'Edit Target' is provided to Edit the Target details.

• The Extracted Attack Objectives from the user provided inputs for Code of Conduct can be seen below. An option to Edit the Code of Conduct is also available.

• Once the user scrolls below, the user can find the Integration Method section. Here, the user can select from the integration method_(HTTP or API)_ of their choice.
• Then, the user clicks on the ‘Add Integration’ button and proceeds to fill out the HTTP integration details(as shown below). Note: HTTP is selected as the default integration method in the application.

• The user can then proceed to fill out the HTTP integration details as mentioned below (using Sample data).

Field Explanation

1. Target Type: Select the type of target (e.g., HTTP).
2. Endpoint: Enter the full API URL that the target system uses for communication. Example: "http://api.example.com/endpoint".
3. HTTP Method: Choose the HTTP method (e.g., POST) used by the target API.
4. Parameters and Headers: Enter any necessary parameters and headers required by the API for successful communication.
5. Request Body: Select the relevant request body type such as JSON, Java Script or plain text from the drop down menu and then Input the relevant structure that the target system expects to receive.
6. Reply Received form Target: Provide a sample or expected response that the target system would generate. Example for JSON Type: .response.
7. Validate Button: After configuring the API and validating the inputs, proceed to the next step. Reset: Reset the configuration if necessary.

• Once all the details have been successfully entered, Click on the ‘Validate’ button as shown above.

• Once the system validates the API successfully, the user can click ‘Save’ and proceed.
• The Target is successfully integrated using the HTTP integration method. The below screen lists the Integration details for the user.

• In the Integration method section, the user can choose the “API” method for integration.

• The API Key can be Generated for the user who wishes to integrate using API integration method.
• Once the integration step is completed, The user gets redirected to the ‘Add New Session’ page where the user can select the ‘Session Type’ of their choice as per their requirement and ‘Start/Run’ their Session.

French Attacker

Fortify now supports multilingual attack generation, starting with Canadian French. This enhancement enables Fortify to generate AI red-teaming attacks directly in the language supported by the target LLM, improving the accuracy of vulnerability detection for multilingual deployments.

Previously, Fortify generated attacks only in English. With this feature, attacks can now be generated natively in Canadian French using a Direct Attack Generation approach, rather than translating English attacks after generation.

If no language is selected, the system defaults to English, ensuring full backward compatibility.

Key Enhancements:

• Native Canadian French attack generation.
• Direct attack generation methodology (no translation pipeline)
• Language selection integrated in Fortify UI.
• Consistent support across all attack-generation modes.
• Minimal architectural changes (prompt-level updates only).
• No changes to Judge or Severity evaluation components.
• Backward compatible with existing English workflows.

---

Agent Red Teaming

Agent Red Teaming lets Fortify test agentic targets for what they do, not just what they say. Instead of only probing a model for harmful text output, Fortify can now attack an agent's actual permitted capabilities — its tools, resources, and actions — to see whether it can be manipulated into unauthorized tool use, data exfiltration, or other policy-violating operations.

Why this exists ?

Traditional red-teaming objectives and methods aren't specific to an agentic target's real-world capabilities. That makes it hard to test whether an agent can be coerced into taking harmful actions rather than simply generating harmful content. Agent Red Teaming closes that gap by introducing objectives that are aware of what an agent can actually do.

This release encompasses agent-relevant objectives into Fortify's existing session types, ahead of a larger, target-specific pipeline. This buys time to properly define the agentic taxonomy before that heavier lift begins.

Agent-Relevant Objectives

Fortify's objective pool now includes an initial set of AI agent attack objectives (approximately 9 to start). These are reconnaissance- and action-oriented objectives — mapped to the OWASP LLM Top 10 — that specifically probe agent behaviors rather than pure content generation. They exist as their own category, separate from the ASI-based objective framework.

These objectives are target-independent— they are not currently tied to a specific target's permitted or denied actions.

Agent-relevant objectives are available across Fortify's existing session types — there is no new "Agentic" session type yet.

What's New?

• The three-artifact pipeline

i. Target Governance Document (A2A card): A schema and authoring flow capturing a target's resources, permitted actions, and denied actions. This is the sourced from the A2A(agent-to-agent) card the client provides for their target. Its the source of truth Fortify uses to understand what an agent is actually allowed to do — and therefore what it means for the agent to be compromised.

ii. Agent-Specific Attack Objectives: Extracted automatically from the governance document (A2A card), and cross-referenced against the Agentic Harm Category Taxonomy and rooted in the OWASP ASI framework. Unlike the target-independent objectives available today, these objectives are tailored to what a given target can actually do.

iii. Vulgaris Session Integration: The objective payload is wired into the existing Vulgaris 1.2 session runner, so Vulgaris can run both multi-turn and single-turn attack sessions directly against the target using target-specific objectives. This phase reuses existing attack methods rather than introducing a new method generator — the attacker and judge run on Fortify's existing Target Specific (Chat Policy) prompts.

• Improved flow

This redesign ensures a modal-based flow, consistent with Fortify's other design patterns, so policy configuration is clearly part of creating a target rather than a surprise afterward.

• New UI labelling

Within the session configuration flow, agent-relevant objectives are marked with a 'clear label/pill' so they're easy to identify and reason about.

Agent-relevant objectives support both:

• Single-turn execution
• Multi-turn execution

• Reporting

No changes to the PDF report format or aggregator logic. Sessions that include agent-relevant objectives render using Fortify's existing report structure, and are judged using Fortify's existing judging pipeline — scoring logic is unchanged in this release.

A Step-by-Step Walkthrough

1. Navigate to 'Target Configuration' tab and click on 'Add New Target' button.(Highlighted in Red)

2. Target Configuration Page has been re-designed with an improved flow. User first enters the 'Target details' and sets-up the 'Integration' method.

• The 'Target Details' section now consists of details such as 'Name of the Target', 'Industry', 'Supported Languages' and an option to 'Select Icon'.
• The 'Integration' section remains the same. Users can select the Integration type i.e. 'HTTP or API' as per their choice.

Note: Helptext is added to every field to assist the users.

3. Once Integration is successfully setup, the user proceeds to enter the 'Fallback Responses' section.

• Please Note: The user needs to fill and save all the mandatory Target details until this point, inorder to proceed to the 'Optional' section availble(Section grayed out here).
• User can skip the Chat/Agent Policy configuration if they wish.

4. Once all the details are saved, the 'Policies' section and the 'Members'[For RBAC Implementation] button is visibile to the users. (See below)

• User clicks on the 'Save' button on the top-right corner of the screen. A 'Success message' is shown depicting successful target creation.

• The target 'Chat Policy' gets opened as default once the Target details are saved.
• User can browse again to the 'Target Configuration' page and see that both the 'Members' button and 'Policies' section is now availble for the user.(Highlighted in Red)

5. Users can 'Edit' their 'Target details' and 'Fallback Response' here before proceeding. Just makes the changes and click on 'Save' to update the details. (Highlighted in Red)

6. Users can now proceed to set Target related policies. In Fortify, we have 2 Policy available i.e. Chat Policy and Agent Policy.

Chat Policy Set Up

7. User can click on the 'Set Up' button in the 'Chat Policy' section to begin Chat policy Configuration. Note: Chat Policy was previously known by the name Code of Conduct(COC). Users gets a brief about what the Chat Policy outlines.
   • The user needs to fill in the following fields to set up the Chat Policy:
   1. Target Purpose and Scope: Define the primary purpose of the target.
   2. Approved Topics: Specify the advanced topics the target is authorized to engage with.
   3. Prohibited Topics and Behaviors: Indicate the topics and behaviors the target should avoid.
   4. Tone of Conversation: Set the tone the target should use.

8. Once the user fills and clicks on 'Save', the Chat Policy configuration gets complete and it triggers Fortify to Extract and generate Objectives as per the Chat Policy. (Highlighted in Red)

• Users can also use the 'Reset' button, incase they want to reset the given inputs. 'Cancel' button

9. Once completed, the users can 'view/edit' their Chat policy along with the Extracted Objectives. (Highlighted in Red) Target version history is also available and can be viewed as applicable. (Highlighted in Green)

10. User can select different 'target version' as deemed fit and complete the Chat Policy set up process.
    • Check the example below, in this target the user can click on the 'Activate Version' button infront of the 'target version' of their choice(if available).
    • Here, two target versions are availble for the users to choose from.

Agent Policy Set Up

11. Similarly, users can now set up their 'Agent Policy' for Red teaming agents by clicking on the 'Set up' button under Agent Policy tile.

Agent Card

12. The system pop-up asks the users to 'Upload your Agent card' to proceed.

• Click on the "Browse file" button or 'drag your file' to upload the Agent Card file. Only ".json" file type is accepted.
• User can click on the 'Learn more about A2A cards' to know more about Agent cards and its related concepts.(Highlighted in Green)
• User can click on 'Get Template' which provides them with a Sample .json example file.(Highlighted in Red) Note: This file needs to saved and edited with the correct inputs inorder to proceed.
• Maximum upload file size is 2MB.

13. An 'Error message' will be shown if the 'file type' or 'maximum upload size' of the file is violated.

14. Similarly, incase of any issues with the 'uploaded Agent card' like 'missing values' or 'empty parameters', the system will throw an 'error message' stating the issue.(Highlighted in Red)

15. Once uploaded successfully, the 'agent card' and its 'file size' appears infront of the file.

16. Users get a quick ability to preview their uploaded 'Agent Card' by clicking on the 'Preview Agent Card' dropdown sign. (Highlighted in Red)

17. Then click on 'Save' button to moving forward, and triggers Fortify to 'Extract Attack Objectives' as per the details mentioned in the 'Agent Card'.

• A 'toast message' is shown after successfully uploading the Agent Card.

• Once the objectives are extracted, it redirects the user to the 'Target Configuration' page. Here the user can see how many objectives have been extracted by clicking on the 'Version History' button. (Highlighted in Red)

18. Users can view their generated 'Agent policy' along with the 'Extracted Objectives'.

• Target version history can be viewed(if applicable).
• User can also click on 'Upload new agent card' button and upload new agent card directly from here .

---