SafePrompt checks incoming messages for attempts to manipulate your AI, with a free plan and self-serve paid plans from $29/month.
Archive note: This page preserves the original launch date and founder quotations. Product descriptions and plan details have been corrected to reflect current published information.
February 20, 2026. SafePrompt announced the general availability of its prompt injection protection API for developers building AI applications. Developers can check incoming text before their model reads it and block messages flagged as unsafe.
SafePrompt gives developers a checkpoint for messages that try to redirect an AI. Prompt injection can arrive through user input or external material the model reads, including retrieved documents and tool results. OWASP's LLM01:2025 guidance describes both direct and indirect injection, with possible consequences including disclosure of sensitive information and unauthorized actions.
SafePrompt checks text before it reaches the model and returns a verdict your application can act on. The response includes a safe boolean, threat details, confidence and reasoning. Your application uses that response to decide whether to pass the text to its AI.
SafePrompt combines pattern detection, external reference detection and AI semantic analysis to identify attempts to compromise the reading AI. Its scope includes instruction overrides, jailbreaks, system-prompt extraction and instructions to exfiltrate data. Developers can also submit retrieved documents and tool results for screening before model consumption.
When deployment session tracking is enabled, session_token requests tracking; it does not guarantee conversation isolation. Its multi-turn tests evaluate submitted payload turns, not gradual escalation coverage. Application permissions and output handling stay in your code, so you can enforce the rules specific to your users and tools.
"Our goal was to make prompt security as simple as Stripe made payments: one API call, transparent pricing, and no gatekeeping."
Ian Ho, Founder, SafePrompt. Excerpt from the original launch statement, February 20, 2026.
Developers can inspect SafePrompt's published benchmark results and try the attack playground before integrating. The playground includes recorded examples and a limited live check for a message you write. Test the API on representative inputs from your own application to assess its fit.
Developers can review SafePrompt's privacy policy for data practices and use the contact page for questions about sharing or retention. The coverage guide explains the screening scope and application controls.
Developers can get a free API key through self-serve signup, with no credit card. The current published plans below replace the original release's plan details. See the pricing page for plan features and billing terms.
SafePrompt provides an HTTP API and JavaScript and Python SDKs. The integration documentation covers authentication, required request fields and response handling. The same core detection pipeline serves all three plans.
"Most developers know they need protection but don't have the security expertise to build it themselves."
Ian Ho, Founder, SafePrompt. Excerpt from the original launch statement, February 20, 2026.
Prompt injection uses input to redirect an AI's behavior or output. It can come directly from a user or indirectly from a document or website the model reads. OWASP's LLM01:2025 guidance describes these attacks and controls that reduce their impact.
SafePrompt checks the text an application submits and returns a safe or unsafe verdict with threat details. Developers place the check before model consumption and use the verdict to block flagged input. The API supports screening user messages, retrieved documents and tool results.
SafePrompt screens incoming text for attempts to compromise the AI reading it, including instruction overrides, jailbreaks, system-prompt extraction, exfiltration instructions and indirect injection in retrieved content. Deployment-enabled session tracking can associate checks without guaranteeing conversation isolation or full-history semantic detection.
SafePrompt's current published Free plan includes 10,000 validations a month, with no credit card. Starter is $29 a month for 500,000 validations, and Business is $99 a month for 1,000,000 validations. Plan features and billing terms are on the pricing page.
SafePrompt provides a prompt injection screening API for developers building chatbots, retrieval applications and AI agents. Founded by Ian Ho, the product helps applications check untrusted text before their AI reads it. Visit safeprompt.dev for product information.