Pangea built security services that developers call through APIs, and in recent years it focused on AI applications. On 16 September 2025, CrowdStrike announced a definitive agreement to acquire Pangea and completed the acquisition on 26 September 2025, to deliver what it calls AI Detection and Response (AIDR) on the Falcon platform. As of 17 October 2026, Pangea's documentation is titled "Documentation for CrowdStrike AI Security Guardrails and Services". Teams searching for Pangea alternatives are usually comparing guardrail APIs, reconsidering after the acquisition, or finding that their agents need more than content guardrails. This guide covers each case. Other vendor guides are on the comparison hub.

What Pangea covers (as of 17 October 2026)

Pangea's documentation lists these services:

  • AI security: AI Guard (LLM guardrails), Prompt Guard (prompt injection) and an AIDR entry for AI visibility.
  • Data protection: Redact, Sanitize, Vault and Secure Share.
  • Threat intelligence: File, IP, Domain, URL and User Intel, plus File Scan and Embargo (export restriction checks).
  • Identity and records: AuthN, AuthZ and Secure Audit Log.

SDKs are listed for C#, Go, Java, Node.js and Python. The docs describe three deployment models (SaaS, Edge and Private Cloud) and two integration options (in-app security and API gateways).

How AI Guard works

The AI Guard quickstart says AI Guard uses detection policies, called recipes, "to identify and block prompt injection, enforce content moderation, redact PII and other sensitive data, detect and disarm malicious content". Each recipe combines one or more detectors. The service accepts plain text or an array of chat messages in a format similar to OpenAI's, and returns sanitized content with a report. Detections are logged, and webhooks can send real-time alerts. The quickstart starts from a free Pangea account.

What the acquisition announcement says

CrowdStrike's release states that Pangea will extend the Falcon platform to the interaction layer where AI is built and used. It claims prompt injection protection "with up to 99% efficacy at sub-30ms latency", and a footnote says the figures come from internal benchmark testing on a GPU-based edge deployment. Ask for your own test on your own traffic before you rely on any guardrail benchmark.

Where Pangea is strong

  • Developer experience. One account, SDKs in five languages and a quickstart that runs in minutes.
  • Breadth of building blocks. Guardrails, redaction, threat intelligence, authentication and an audit log service come from one API family.
  • Deployment choice. SaaS, edge and private cloud options cover teams with data-location limits.
  • A large parent. For a CrowdStrike customer, AIDR on Falcon may fit existing operations and contracts.

Why teams evaluate alternatives

  1. Not a CrowdStrike customer. After an acquisition, roadmap and packaging follow the parent. Confirm both with the vendor.
  2. A different guardrail engine. Detection quality varies by language, domain and attack style. Teams often test two or three engines on their own data.
  3. Agents that act. A content guardrail inspects text going in and out of a model. Deciding whether a given agent may call a given tool with given arguments is a separate authorization decision; ask whether the vendor's authorization service covers it.
  4. Evidence for auditors. Some buyers want signed decision records they can verify offline.

Alternatives by category

What you need Category Examples covered on this site
A guardrail API for prompts and responses Guardrail API Pangea, Lakera
Guardrails as part of a wider AI security platform AI security platform Prisma AIRS, Cisco AI Defense
Agent-focused testing and runtime protection Agent-native security Lasso Security, Prompt Security
Guardrails in an AI gateway you already route through AI gateway Portkey, Cloudflare AI Gateway
Agent identity, tool-call policy, approvals and signed evidence Agent control plane Praesidia

Our post on content guardrails for AI agents explains where content checks sit in an agent's request path.

Where Praesidia fits

Praesidia is our product. It is an AI agent control plane. It has guardrails, but it is not a drop-in replacement for a guardrail API called from inside your application. What it does:

  • Guardrails in the model gateway. Personal data, payment cards, secrets, prompt injection and jailbreaks are checked on the way in and the way out. Rule matches can block, warn, redact or replace; ML and LLM guardrails block or warn. ML guardrails are on the Team plan and LLM guardrails on its Advanced tier. The AI gateway guide explains where the model gateway sits and what it enforces.
  • Identity per agent. Each registered app, agent and MCP server gets its own credentials for authentication and attribution.
  • A decision on each governed tool call. The decision uses the agent's identity, the tool's permission on that connection and the arguments. New policies start in observe mode; denials and holds apply once a policy is set to enforce.
  • Approvals. A step-up rule turns an MCP tool call into a durable approval request that expires if nobody answers.
  • Signed record. Praesidia records the decisions it makes in an append-only, hash-chained audit trail, signed under the default configuration; an operator can turn signing off. A signed record shows it was not changed afterwards. It does not show that every action was captured. Calls that do not pass through Praesidia are not recorded.

Guardrails in Praesidia apply to traffic routed through its gateway. A team that wants to call a guardrail as a function inside its own code is better served by a guardrail API. The guardrails documentation lists the detectors and actions.

Questions to ask in an evaluation

  1. Where does the check run? In your code through an SDK, in a gateway, or both. Each choice changes what traffic is covered.
  2. What happens on a hit? Block, redact, or report only. Confirm the default for each detector.
  3. How was the benchmark run? Ask for the test set, the hardware and the false-positive rate, not only the detection rate.
  4. Do you need decisions on actions? If agents call tools, ask who decides each call and who can hold it for a person.
  5. What changes after the acquisition? Ask about packaging, pricing and support for the standalone APIs.

Common questions

Who owns Pangea? CrowdStrike announced the acquisition on 16 September 2025 and completed it on 26 September 2025. As of 17 October 2026, Pangea's documentation is branded as CrowdStrike AI Security Guardrails and Services.

What is Pangea AI Guard? AI Guard is Pangea's LLM guardrail service. It applies recipes of detectors to prompts and responses to block prompt injection, moderate content, redact sensitive data and disarm malicious content.

Is a guardrail API enough for AI agents? For chat-style applications it can be. Agents that call tools also need a decision on each action: which agent, which tool, which arguments. That is an authorization question, which content inspection does not answer.

Does Praesidia replace Pangea? Not as an in-app guardrail API. Praesidia applies guardrails to traffic through its model gateway and adds per-agent identity, tool-call decisions, approvals and signed records.

Sources (as of 17 October 2026)