Prisma Browser Expands Agentic Security to Desktop and Browser Extension Agents
AI agents are rapidly transforming enterprise productivity, but because they often use the same authenticated sessions as employees, they create a massive blind spot for security teams. Securing AI agents in the browser requires visibility into their activities, granular access controls, and protection against unauthorized actions and sensitive data exposure.
As we outlined in our previous blog, Where Agentic AI Meets Enterprise-Grade Security, Prisma Browser already provides foundational protection for these autonomous workflows. Now, Palo Alto Networks is extending that protection to desktop and browser extension-based agents, ensuring you can unleash AI productivity without losing control of your sensitive data.
Agentic Browsing Introduces New Security Risks
When agents act on behalf of employees, they may access sensitive records, modify information, or execute transactions beyond their intended tasks.
This introduces critical security challenges:
- Identity blind spots: Difficulty distinguishing employee actions from agent activity.
- Uncontrolled actions: Agents operating beyond intended permissions.
- Sensitive data exposure: Agents accessing or sharing confidential information.
- Agent hijacking: Malicious web content manipulating agent behavior.
- Accountability gaps: Limited visibility and auditability of agent actions.
Securing the AI model alone isn't enough. Organizations need security where agents interact with enterprise applications.
Expanding AI Agent Security Across Browser Environments
AI agents access the browser in different ways. Prisma Browser now extends its existing security capabilities across these scenarios:
- Desktop agents: Secure agents interacting with web applications through Prisma Browser's embedded MCP server.
- Embedded browsers within desktop agents: Extend security to agents using their own browsers to access web applications.
- Extension-based agents: Securing agentic workflows triggered by agentic extensions that transform any browser into an agentic browser.
- In-browser agents: Secure AI-assisted workflows powered by Prisma Browser's embedded agent.
This expanded coverage builds on five core Prisma Browser security capabilities.
1. See AI Activity: Know Who Did What
When employees and agents share an authenticated session, traditional security tools may struggle to distinguish their actions. Prisma Browser provides detailed visibility into agentic activity, including websites visited, actions performed, the employee represented, and policy violations.
Activity is recorded in security event logs with supporting evidence, enabling investigation and accountability. Organizations can distinguish what employees do from what agents do, even within the same browsing session.
2. Govern Agentic Actions: Different Policies for Humans and Agents
Just because an agent acts on behalf of an employee doesn't mean it should have the same privileges. Prisma Browser enables organizations to enforce different policies for humans and agents, even when both interact with the same web page. By applying least-privilege access controls to AI agents, organizations can limit autonomous actions without restricting employees' authorized activities.
Organizations can:
- Control web access: Allow employees to access websites while restricting or blocking agents.
- Enforce different permissions: Give employees full access while limiting agents to read-only actions on the same page.
- Restrict sensitive actions: Prevent agents from completing login forms or performing unauthorized activities.
- Require human approval: Apply human-in-the-loop workflows and step-up MFA for sensitive operations.
These capabilities extend Prisma Browser's existing granular security controls to agents, enabling independent governance within the same browsing session.
3. Protect Enterprise Data: Extend DLP to Agentic Activity
AI agents interact with sensitive enterprise information, from customer records to financial documents and intellectual property.
Prisma Browser leverages Palo Alto Networks Enterprise DLP's advanced data classification capabilities with over 1000 classifiers to accurately identify sensitive information and enforce precise, context-aware protection policies for both humans and agents.
Key capabilities include:
- Sensitive data masking: Hide confidential fields or entire pages from agents while keeping them visible to authorized employees.
- Granular data access controls: Apply different data access permissions to humans and agents, even on the same web page.
- Last-mile data controls: Inspect and protect sensitive information before an agent types it into a web application.
An employee can view and modify a customer record while an agent working on the same page sees only permitted information and has read-only access.
4. Stop AI Threats: Protect Against Agent Hijacking and Web Attacks
Prompt injection attacks, including malicious instructions embedded in web content, can manipulate agents into revealing sensitive information or performing unauthorized actions.
Prisma Browser leverages Prisma AIRS, the industry’s most comprehensive AI security platform, to protect AI interactions in real time and govern AI agents at scale. Going beyond basic prompt filtering, Prisma AIRS delivers advanced protection against more than 30 types of prompt injection attacks across 14 languages, with detection of malicious URLs, code-based threats, and destructive actions. It also identifies over 1,000 sensitive data patterns and enforces customizable topic, toxicity, and content guardrails. By analyzing prompts and content in real time, Prisma Browser helps stop compromised agents before malicious actions can execute.
Agents also face traditional web threats. Prisma Browser provides protection through Palo Alto Networks' advanced security engines:
- Advanced Web Protection: AI-powered detection of malicious web activity.
- Advanced URL Filtering and WildFire: Protection against phishing, malicious URLs, and malware that blocks up to 15M new threats every day.
- Advanced Browser Protection: Patented mitigation against browser exploits and zero-day attacks.
- Advanced Extension Security: Detection of malicious browser extension behavior.
Together, these capabilities protect agentic workflows against both traditional web threats and emerging AI-specific attacks.
5. Safeguard Credentials: Allow Agents to Sign In Without Exposing Passwords
To complete key tasks, AI agents frequently need to authenticate into enterprise applications. Common workarounds, such as pasting credentials into prompts, sharing session tokens, or hardcoding secrets, expose sensitive assets to risk if an agent is manipulated.
Prisma Browser integrates secure credential management, powered by Idira, enabling agents to authenticate safely while ensuring credentials remain fully protected under enterprise control.
Key credential governance features include:
- Secure authentication: Grant agents access to required application accounts using credentials managed entirely within the password manager.
- Prompt-isolated credentials: Prevent secrets from entering agent context or instructions, eliminating credential leakage from prompt injection or hijacked agents.
- Granular access policies: Control agent authentication permissions on a per-application basis using agent-specific security policies.
- Step-up verification for sensitive accounts: Require human approval or multi-factor authentication before an agent accesses privileged or high-risk accounts.
For example, when an employee requests an agent to retrieve an invoice from a supplier portal, the agent completes sign-in via the password manager without exposure to the underlying password.
Accelerate Agentic Productivity Without Compromising Security
Imagine an employee asking an agent to prepare a customer report. The agent accesses Salesforce, retrieves information, reviews documents, and completes a form.
Prisma Browser enables this workflow while maintaining visibility, enforcing agent-specific permissions, protecting sensitive data, and requiring human approval when necessary.
With expanded coverage for desktop and extension-based agents, organizations can apply consistent security across more agentic workflows without separate approaches for each agent type. As AI agents become integral to enterprise operations, security must keep pace with how they interact with the web. Regardless of how agents access the browser, Prisma Browser helps organizations secure what they do there.
Discover how Prisma Browser helps organizations secure desktop and browser-based AI agents with visibility, granular access controls, data protection, and threat prevention. Schedule a demo.