Enterprise AI and Agent Security & Control

AI tools (ChatGPT, Claude, Gemini, and others) are already being used by employees, and AI functionality is actively integrated into applications and services.

Meanwhile, organizations often lack:

• sufficient visibility of how AI is used;
• control over what data is transmitted to AI services;
• mechanisms for identifying risky interactions;
• necessary context for incident investigation;
• a coordinated approach between SecOps, IT and engineering teams.

Falcon AI Detection and Response (AIDR) provides visibility and control over AI interactions without the need to restrict the use of artificial intelligence.

Control the use of AI by employees in a managed environment.

Monitor AI activity in managed browsers using Browser Collector AIDR, a lightweight browser extension for managed endpoints, ensuring the collection and analysis of employee interactions with web-based AI tools followed by the application of security policies.

Features

• visibility of employee use of AI tools;
• analysis of interactions (prompts and responses);
• detection of sensitive data transfer (PII, credentials, etc.);
• enforcement of control policies;
• support for data protection scenarios in GenAI.

When it’s relevant

• employees using public AI tools;
• requirements for data transfer control;
• need for visibility without a complete ban on AI.

Control and Security of AI Activity in Applications and Agent Environments

Monitoring and controlling AI activity in applications, services, and agent environments using a set of AIDR collectors, which provide the collection and analysis of AI telemetry at various levels of infrastructure and the application of security policies.

Collectors

• Browser – a browser extension that records user interactions with AI provider sites (ChatGPT, Claude, Gemini, and others) in managed browsers;
• Application — integration through SDK and API for instrumentation of internal applications with built-in (inline) AI security checks;
• Agentic — MCP (Model Context Protocol) proxy that records AI traffic between MCP clients and servers;
• Gateway — integration at the network proxy level (Kong, LiteLLM, Portkey, and others) that inspects AI traffic at the API gateway level;
• Cloud — integration with cloud platforms that ingests logs and events related to AI from supported platforms (AWS Bedrock);
• OpenTelemetry — standardized telemetry instrumentation for collecting data related to AI from applications and services.

Capabilities

• visibility of AI activity in applications and services;
• detection of risk scenarios (e.g., prompt injection);
• application of policies to AI interactions;
• ensuring compliance and conformity with regulatory requirements.

When it’s Relevant

• AI is integrated into products or internal systems;
• control of interactions at the API or application level is needed;
• there are risks associated with the exploitation of AI logic.

AIDR collects AI telemetry, analyzes interactions, and enforces security policies.

What data is collected:

• prompts;
• AI responses;
• metadata (user, device, application, timestamp, etc.);
• analysis results (identified risks, actions taken).

Types of policies

Access rules — based on attributes (user, device, application).
Prompt rules — content analysis of prompts and responses.

What risks AIDR helps to reduce

• transmission of sensitive data to AI services;
• use of unauthorized AI tools;
• lack of visibility into AI usage;
• risky or dangerous interactions with AI;
• violations of internal policies and compliance.

AIDR does not replace existing security tools (SIEM, DLP, EDR), but complements them by adding visibility and control specifically for AI interactions.

CrowdStrike collaborates with OpenAI as part of the Trusted Access for Cyber (TAC) program, gaining access to specialized cybersecurity models.
This allows:
• using AI to detect complex attacks
• analyzing threats in real time
• improving the accuracy of security solutions