Gray Swan Cygnal

What it does

Cygnal is Gray Swan's runtime guard for AI applications. It sits inline between users and a model, between an agent and the tools it calls, and between a retrieval pipeline and the content it returns, classifying each prompt, response, and tool call for prompt injection, jailbreaks, unsafe outputs, and risky agent actions.

Developers integrate it by pointing an OpenAI-compatible client at the Cygnal base URL and adding a Gray Swan API key header, while keeping their own model provider key. When a policy is violated, Cygnal returns a refusal message and sets finish_reason to violation. Every classification is logged to the Gray Swan dashboard. The product page says policy thresholds, action types such as block, flag, or rewrite, and routing logic are configurable.

It suits enterprise teams running customer-facing chat, internal copilots, or tool-using agents in production. Access starts with a Gray Swan Platform account to obtain an API key, and the website also offers demo booking. Pricing is not published, so buyers should confirm cost with the vendor. Cygnal is a runtime control and not a pre-release testing tool; that role belongs to Shade.

What it can do

• Classifies direct and indirect prompt injection, including content delivered through retrieved documents and tool output
• Detects jailbreak attempts aimed at bypassing safety training or system guardrails
• Flags or blocks unsafe model outputs against configured policies
• Identifies agentic risk patterns such as unauthorized tool use and scope violations
• Logs every classification to the Gray Swan dashboard for monitoring and review
• Supports custom policies defined in natural language and configurable block, flag, or rewrite actions

Works with

• OpenAI
• Anthropic
• Google Gemini
• OpenRouter

Before you choose

• Cygnal is a runtime guard; for pre-release adversarial testing the vendor points to Shade, a separate product.
• Pricing and plan details are not published, so confirm costs with Gray Swan before committing.
• Applications must handle violation responses, where finish_reason is violation and a refusal message is returned.

Listed under

• Security
• AI and agent security

Get started

Docs instruct setting GRAYSWAN_API_KEY after signing up; a demo can also be booked.

1. Sign up on the Gray Swan Platform and request an API key, then set it as the GRAYSWAN_API_KEY environment variable.
2. Set the API key for your chosen model provider (for example OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, or OPENROUTER_API_KEY).
3. Change the client base_url to https://api.grayswan.ai/cygnal and add the grayswan-api-key header; check finish_reason for violation responses.

No price, plan, trial, or usage rate is published for Cygnal. Contact the vendor for current terms.

We couldn’t find published prices for Gray Swan Cygnal.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Cygnal - Runtime Monitoring and Protection ↗www.grayswan.ai
• Quickstart | Gray Swan Docs ↗docs.grayswan.ai
• Gray Swan - Enterprise Security for AI-Powered Applications ↗www.grayswan.ai
• Shade - Adversarial Red-Teaming ↗www.grayswan.ai
• Arena - Red-Teaming Battlefield for Breaking AI Models ↗www.grayswan.ai

Company website

grayswan.ai: product, pricing and policy pages

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.