Academic Security Findings in the AI Models Table

Note: This feature is only available with a Premium AI subscription.

Overview

This feature brings academic research on AI model vulnerabilities directly into your AI Models table. Here’s what you get and why it matters:

• Immediate Security Context: See at a glance which AI models in your environment have known vulnerabilities, based on the latest academic research.

• Aggregated Risk Indicators: Each model displays a count of findings, with severity indicators, so you can quickly spot high-risk models.

• Detailed Vulnerability Insights: For each finding, you’ll see:

  • Vulnerability classification

  • Attack type and affected models

  • Vulnerability score and severity.

  • References to research papers and academic sources

  • Technical details, attack vectors, and mitigation advice

• Actionable Recommendations: Each finding includes remediation steps, helping you proactively secure your AI infrastructure.

• No More Context Switching: All this information is available right where you already manage your AI models—no need to jump between different tools or interfaces.

Why it’s valuable:
You can assess and address AI model risks faster, make informed security decisions, and stay ahead of emerging threats—without leaving your familiar workflow.

Getting it done

Step 1: Open the AI Models Table

• Navigate to the AI Models section in your product dashboard.

Step 2: Review Security Findings at a Glance

• In the table, each model now displays an aggregated count of security findings, with severity icons (e.g., red for critical, yellow for moderate).

Step 3: Dive Deeper with the Side Panel

• Click on any model row to open its side panel.

• You’ll see a new AI Models Findings tab. Click it to view a list of all academic findings associated with that model.

  

Step 4: Explore Detailed Information

Click on any finding in the list to open a detailed view.
Here you’ll find:

1. Security Overview - Contains the severity and ID of the vulnerability.

2. Finding Information - Contains the description and mitigation steps, based on the academic papers.

3. CVSS Score - Contains information about the vulnerability classification and attack vectors.
   Note: CVSS scores for ML models are determined using a hybrid approach combining LLM-based assessment with manual supervision by Mend AI analysts. The scoring process was initially set up using CVSS 4.0 framework, later converted to CVSS 3.x.

4. References - Contains links to the academic research papers.

Step 5: Take Action

• Use the provided recommendations to address vulnerabilities.

• Reference the academic sources for deeper understanding or compliance documentation.

Step 6: Stay Updated

• As new security findings are published, the table and findings tab will update automatically—no manual refresh needed.

Limitations & Notes

• Data Migration & Rescanning: Findings are based on the current scan of your environment. If you migrate data or rescan, findings may update or change.

• Academic Source Scope: Currently, findings are based on academic research papers. Future updates may include additional sources.

• Academic Security findings currently cannot be used in Automation Workflows.

• Research papers don't always provide specific action items to mitigate risks.
  Expect Mend AI to continuously update and improve the information about mitigation and mitigation steps.