Pangea builds security guardrails and security APIs for teams that develop AI applications. It is for developers who need to keep sensitive data and attacks out of the prompts, responses and pipelines their applications handle, and who would rather call a service than build and run that security themselves.
The platform covers both sides of that job: filtering what goes into and comes out of large language models, and the authentication, secrets and audit basics any application needs. It also reaches AI agents through the tools they call.
At A Glance
| Vendor | Pangea |
|---|---|
| Category | Security guardrails and security APIs for AI applications |
| Identities covered | Application users, secrets and keys, AI agents calling MCP tools |
| Deployment | SaaS, Edge (customer hosted data processing) or Private Cloud |
| Agentic AI | Open source MCP Proxy and MCP Server, AuthN for FastMCP |
| Ownership | CrowdStrike |
| Market classification | Agentic AI Identity |
What Pangea Does
Most AI applications pass user input, retrieved documents and model output through several steps, and any of those steps can leak personal data or carry an attack. Pangea puts security controls into that path as services the application calls, rather than as separate infrastructure the team has to run. Because the controls are ordinary API calls, they sit inside the application’s own logic, can be applied at each step where data changes hands, and leave a record of what was caught. Developers pick only the services they need and decide whether Pangea or their own cloud processes the data.
Products
AI Guard
AI Guard is the guardrail service for large language models. It removes personal data, other sensitive data and malicious content from ingestion pipelines, prompts and responses before they reach the model or the user. A default configuration ships with recipes for common situations, and teams can define their own entities to be reported, removed or encrypted. Every detection is written to an audit trail, and webhooks can raise alerts in real time.
Prompt Guard And AIDR
Prompt Guard detects and blocks prompt injection, the attack where crafted input tries to override a model’s instructions. AIDR, short for AI detection and response, gives security teams visibility of how AI is being used and insight into what the guardrails are catching.
Identity And Secrets Services
These services cover the identity plumbing of an application, so developers do not have to build it themselves.
| AuthN | Login and user management, with React Auth and Vanilla JS libraries for the front end |
|---|---|
| AuthZ | Authorisation decisions inside the application |
| Vault | Storage and protection for secrets and keys |
| User Intel | Checks against credential breach data |
Data Protection And Audit
Secure Audit Log keeps a tamper resistant record of events, with an Audit Log Viewer library for showing it in an application. Redact and Sanitize remove sensitive information and malicious URLs from content, Secure Share handles authenticated file transfer, and Embargo checks digital export restrictions.
Threat Intelligence
File Intel, File Scan, IP Intel, Domain Intel and URL Intel let an application test files, domains and URLs for malicious content and look up the reputation and location of an IP address.
How It Works
Pangea offers three deployment models.
| Component | What it does |
|---|---|
| Pangea SaaS | Pangea hosts and manages everything, and data is processed in Pangea Cloud. |
| Edge | The control plane stays hosted by Pangea, while data processing runs in the customer’s own AWS, Azure or GCP environment. Configuration is still managed from the hosted console. |
| Private Cloud | Both the data plane and the control plane run inside the customer’s infrastructure and are managed by the customer. |
| Calling the services | Applications use the services in app or through API gateways, with SDKs for several languages or the REST APIs directly. Detection events can be pushed to other systems with webhooks. |
Identities And Systems Covered
| Application users | Sign in and user management through AuthN, access decisions through AuthZ |
|---|---|
| Secrets and keys | Stored in Vault, and detected in text by AI Guard |
| Breached credentials | Checked with User Intel |
| AI agents and MCP servers | Guarded by the MCP Proxy, with AuthN available for FastMCP servers |
Agentic AI Capabilities
Pangea MCP Proxy
The MCP Proxy is an open source wrapper for existing MCP servers. It passes “only AI Guard sanitized inputs to the underlying tool” and sends only sanitised output back to the agent, so an agent’s tool calls are filtered without changes to the agent or the MCP server.
Pangea MCP Server
The open source MCP Server works the other way round: it exposes Pangea’s guardrail services as tools, so an agent can call them directly.
AuthN can also be added to FastMCP servers, so an MCP server authenticates the users and agents calling it.
Company Background
| Founder and CEO | Oliver Friedrichs |
|---|---|
| Ownership | CrowdStrike announced a definitive agreement to acquire Pangea on 16 September 2025 |
| Research | Studies of prompt injection, including an analysis of more than 300,000 prompt injection attempts |
Note: This overview is compiled from Pangea’s publicly available information as of 4 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.