AI Red Teaming
Focus
Focus
Prisma AIRS

AI Red Teaming

Table of Contents

AI Red Teaming

Learn about new features for Prisma AIRS AI Red Teaming.
Here are the new Prisma AIRS AI Red Teaming features.

Multi-turn Attack Support

February 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
Multi-turn attack support enables you to conduct sophisticated AI Red Teaming scenarios by maintaining conversation context across multiple interactions with your target language models, extending beyond basic single-turn testing to simulate realistic attack patterns where adversaries gradually manipulate model behavior through sequential prompts that mimic how actual users interact with AI systems in production environments.
When you configure multi-turn attacks, AI Red Teaming automatically manages conversation state, supporting both stateless APIs that use past conversations and stateful backends that maintain session context internally.
  • For stateless configurations, you specify the assistant role name used by your target API, and the system automatically builds and transmits the complete conversation history with each subsequent request.
  • For stateful configurations, you define how session identifiers are extracted from responses and injected into follow-up requests, allowing the system to maintain context without resending entire conversation histories.
AI Red Teaming supports all major LLM providers including OpenAI, Gemini, Bedrock, and Hugging Face while also accommodating custom endpoints through flexible configuration options that auto-detect common message formats.
You can leverage multi-turn attacks to test sophisticated vulnerabilities that only emerge through extended conversations with AI agents, particularly when evaluating chatbots and virtual assistants that maintain memory across interactions to verify whether context from earlier turns can be exploited to bypass safety controls in later exchanges. If you are assessing custom API endpoints or self-hosted LLM backends, multi-turn testing reveals how these systems handle conversation state management and whether attackers could poison agent memory or create context confusion, while organizations migrating from one LLM provider to another can use multi-turn campaigns to ensure their new infrastructure maintains the same security posture across extended interactions.

Agentic Target Profiling

February 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
Target Profiling enhances your AI security assessments by gathering comprehensive contextual information about your AI endpoints, enabling more accurate and relevant vulnerability discoveries. With Target Profiling, you can leverage both user-provided information and dynamic Agentic Profiling to get detailed profiles of your AI models, applications, and agents.
Target Profiling collects critical information about your AI systems, including industry context, use cases, competitive landscape, and technical foundations such as base models, architecture patterns, and accessibility requirements. AI Red Teaming's Agentic Profiling capability connects with and interrogates your target endpoint to discover all the business and technical context of the target. This automated approach saves you time while ensuring comprehensive coverage of contextual factors that influence security risks.
The feature provides you with a centralized view where you can visualize all gathered context. You can distinguish between user-provided information and agent-discovered data, giving you full transparency into how your target profiles are constructed.

Enhanced AI Red Teaming for AI Agents and Multi-Agent Systems

February 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
You can now leverage Prisma AIRS AI Red Teaming's enhanced capabilities to comprehensively assess the security posture of your autonomous AI agents and multi-agent systems. As your organization deploys agentic systems that extend beyond traditional AI applications to include tool calling, instruction execution, and system interactions, you face an expanded and more complex attack surface that requires specialized security assessment approaches. This advanced AI Red Teaming solution addresses the unique vulnerabilities Inherent in generative AI agents by employing agent-led testing methodologies that craft targeted goals and attacks specifically designed to exploit agentic system weaknesses.
When you configure your AI Red Teaming assessments, the system automatically tailors its approach based on your target endpoint type, enabling you to uncover critical vulnerabilities such as tool misuse where malicious actors manipulate your AI agents to abuse their integrated tools through deceptive prompts while operating within authorized permissions. The solution also identifies intent breaking and goal manipulation vulnerabilities where attackers redirect your agent's objectives and reasoning to perform unintended tasks.
This targeted approach ensures you can confidently deploy AI agents in production environments while maintaining robust security controls against the evolving threat landscape targeting autonomous AI systems.

Secure AI Red Teaming with Network Channels

January 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
Network Channels is a secure connection solution that enables AI Red Teaming to safely access and analyze your internal endpoints without requiring firewall rules for specific IP addresses or opening inbound ports. This enterprise-grade solution puts you in complete control of the connection, allowing you to initiate and terminate access while maintaining your security perimeter.
The Network Channels enables you to conduct secure, continuous AI Red Teaming assessments against user APIs and models hosted within private infrastructure. Network channels eliminates the need for users to expose inbound ports or modify firewall configurations, adhering strictly to Zero Trust principles.
A channel is a unique communication pathway that clients use to establish connections. Each channel has a unique connection URL with auth credentials. You will need to create and validate a channel first, before using it to add a target. Multiple channels can be created for different environments and each channel can handle multiple targets accessible to it.
The solution utilizes a lightweight Network Channels client deployed within the user’s environment. This client establishes a persistent, secure outbound WebSocket connection to the Palo Alto Networks environment, facilitating seamless testing of internal systems without the risks associated with allowing specific IP addresses through your firewall or inbound access.
Additionally, you will be provided with a docker pull secret from Strata Cloud Manager, which you can use to pull the docker image and helm chart for the network channels client.
This combined solution is ideal for:
  • Restricted Environments: Conducting assessments for enterprise users with air-gapped systems or strict compliance requirements.
  • Continuous Monitoring: Maintaining reliable, persistent connectivity for real-time AI security updates.
  • Automated Workflows: Deploying network broker clients across distributed infrastructure using existing container orchestration (Kubernetes/Helm) without manual intervention.
Key Benefits
  • Enhanced Security: No need to expose internal endpoints or modify firewall rules.
  • Complete Control: Initiate and terminate connections on demand.
  • Easy Setup: Simple client installation process.
  • Flexible Management: Create and manage multiple secure channels for different environments.
  • Reusability: Use the same connection for multiple targets.
  • Enterprise Ready: Designed for organizations with strict security requirements.

Remediation Recommendations for AI Red Teaming Risk Assessment

January 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
The Recommendations feature enables you to seamlessly transition from identifying AI system vulnerabilities through Red Teaming assessments to implementing targeted security controls that address your specific risks. This feature closes the critical gap between AI risk assessment and mitigation by transforming vulnerability findings into actionable remediation plans. The remediation recommendations can be found in all Attack Library and Agent Scan Reports.
When you conduct AI Red Teaming evaluations on your AI models, applications, or agents, this integrated solution automatically analyzes the discovered security, safety, brand reputation, and compliance risks to generate contextual remediation recommendations that directly address your specific vulnerabilities.
The generated contextual remediation recommendations include two distinct components:
  • Runtime Security Policy configuration: Rather than configuring runtime security policies through trial and error, you receive intelligent guidance that maps each identified risk category to appropriate guardrail configurations, such as enabling prompt injection protection for security vulnerabilities or toxic content moderation for safety concerns.
  • Other recommended measures: The system identifies successfully compromised vulnerabilities, and provides the corresponding remediation measures by prioritizing them based on effectiveness and implementation feasibility, allowing you to eliminate manual evaluation and focus resources on high-impact fixes.
For organizations deploying AI systems in production environments, this capability ensures that your runtime security configurations and remediation measures are informed by actual risk insights rather than generic best practices, resulting in more effective protection against the specific threats your AI systems face.
The remediation recommendations appear directly in your AI Red Teaming scan reports, providing actionable guidance. You can then manually create and attach the recommended security profiles to your desired workloads, transforming AI risk management from a reactive process into a proactive workflow that connects vulnerability discovery with targeted protection.

AI Red Teaming Executive Reports

January 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
When you conduct AI Red Teaming assessments, you often need to share results with executive stakeholders like CISOs and CIOs who require high-level insights rather than granular technical details. The exportable executive PDF report feature addresses this need by transforming your AI Red Teaming assessment data into a consumable format. You can now generate comprehensive PDF reports that consolidate the essential information from the web interface, organized to highlight critical takeaways and strategic insights in a format that can be shared easily with executives. While security practitioners can leverage CSV and JSON export formats to access detailed findings for remediation purposes, this PDF format is highly valuable for CXOs who intend to understand high level risk assessment of any AI system.
Use this feature to communicate security posture and risk assessments to executive leadership who may not have the time or technical background to parse through detailed CSV exports or navigate complex web interfaces. The PDF format ensures you can easily distribute reports through email or include them in executive briefings and board presentations. The report contains AI summary, key risk breakdown, high-level overview charts, and metrics that matter most to decision-makers.
You can use this feature when preparing for executive reviews, board meetings, or any scenario where you need to demonstrate the effectiveness of your security controls and communicate risk exposure to non-technical stakeholders. The structured format with AI summary, overview sections, and detailed attack tables, ensures that both strategic insights and supporting technical evidence are readily accessible, enabling informed decision-making at the executive level.

AI Summary for AI Red Teaming Scans

January 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
When you complete an AI Red Teaming scan, you receive an AI Summary (in the scan report) that synthesizes key risks and their implications. This executive summary eliminates the need for manual interpretation of technical data, allowing you to quickly understand which attack categories or techniques pose the greatest threats to your systems and what the potential business impact might be.
The AI Summary contains the scan configuration, key risks, and implications.
This capability is particularly valuable when you need to communicate AI risk assessment results across different organizational levels or when preparing briefings for leadership meetings. Rather than struggling to translate technical vulnerability reports into business language, you can rely on the AI Red Teaming generated report to articulate security, safety, compliance, brand, and business risks in terms that resonate with executive audiences. This summary is also valuable in prioritising remediation measures which teams can adopt for a safer deployment of AI systems in production.

Enhanced AI Red Teaming with Brand Reputation Risk Detection

January 2026
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
You can now assess and protect your AI systems against Brand reputation risks using Prisma AIRS enhanced AI Red Teaming capabilities. This feature addresses a critical gap in AI security by identifying vulnerabilities that could damage your organization's reputation when AI systems interact with users in production environments. Beyond the existing Security, Safety, and Compliance risk categories, you can now scan for threats including Competitor Endorsements, Brand Tarnishing Content, Discriminating Claims, and Political Endorsements.
The enhanced agent assessment capabilities automatically generate goals focused on Brand Reputational risk scenarios that could expose your organization to public relations challenges or regulatory scrutiny. You benefit from specialized evaluation methods designed to detect subtle forms of reputational risk, including false claims and inappropriate endorsements that traditional security scanning might miss. This comprehensive approach allows you to proactively identify and address potential brand vulnerabilities before deploying AI systems to production environments, protecting both your technical infrastructure and corporate reputation in an increasingly AI-driven business landscape.

Error Logs and Partial Scan Reports

December 2025
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
When you conduct AI Red Teaming scans using Prisma AIRS, you may encounter situations where scans fail completely or complete only partially due to target system issues or connectivity problems. The Error Logs and Partial Scan Reports feature provides you with comprehensive visibility into scan failures and enables you to generate actionable reports even when your scans don't complete successfully. You can access detailed error logs directly within the scan interface, both during active scans on the progress page and after completion in the scan logs section, allowing you to quickly identify whether issues stem from your target AI system or the Prisma AIRS platform itself.
This feature particularly benefits you when conducting Red Teaming assessments against enterprise AI systems that may have intermittent availability or response issues. When your scan completes the full simulation but doesn’t receive valid responses for all attacks, AI Red Teaming marks it as partially complete rather than failed. You can then choose to generate a comprehensive report based on the available test results, giving you valuable security insights even from incomplete assessments. AI Red Teaming transparently informs you about credit consumption before report generation and clearly marks any generated reports as partial scans, indicating the percentage of attacks that received responses.
By leveraging this capability, you can maximize the value of your Red Teaming efforts, troubleshoot scanning issues more effectively, and maintain continuous security assessment workflows even when facing target system limitations or temporary connectivity challenges during your AI security evaluations.

Automated AI Red Teaming

October 2025
Supported for:
  • Prisma AIRS (Managed by Strata Cloud Manager)
Palo Alto Networks' is an automated solution designed to scan any AI system—including LLMs and LLM-powered applications—for safety and security vulnerabilities.
The tool performs a Scan against a specified Target (model, application, or agent) by sending carefully crafted attack prompts to simulate real-world threats. The findings are compiled into a comprehensive Scan Report that includes an overall Risk Score (ranging from 0 to 100), indicating the system's susceptibility to attacks.
Prisma AIRS offers three distinct scanning modes for thorough assessment:
  1. Attack Library Scan: Uses a curated, proprietary library of predefined attack scenarios, categorized by Security (e.g., Prompt Injection, Jailbreak), Safety (e.g., Bias, Cybercrime), and Compliance (e.g., OWASP LLM Top 10).
  2. Agent Scan: Utilizes a dynamic LLM attacker to generate and adapt attacks in real-time, enabling full-spectrum Black-box, Grey-box, and White-box testing.
  3. Custom Attack Scan: Allows users to upload and execute their own custom prompt sets alongside the built-in library.
A key feature of the service is its single-tenant deployment model, which ensures complete isolation of compute resources and data for enhanced security and privacy.