Zapier Launches AI Guardrails

Zapier, providers of an anrtificial intelligence orchestration platform, has launched AI Guardrails, a set of builder-added safety checks that run directly inside automated workflows.

AI Guardrails lets teams detect personally identifiable information (PII), identify prompt injection attempts, and flag toxic or harmful content before AI outputs ever touch a CRM, database, or customer inbox. It embeds real-time safety checks directly into Zaps, Agents, and MCP-connected tools.

"Every company using AI in production has the same question: how do we know the outputs are clean before they hit our systems?" said Brandon Sammut, chief people and AI transformation officer of Zapier, in a statement. "AI Guardrails gives teams an actual enforcement layer, not a policy document sitting in a shared drive somewhere. It runs inline, in production, on every single workflow that needs it."

AI Guardrails allows users to add a safety step directly into any workflow. After an AI model generates output, the guardrail checks it against the selected detection type and returns structured results. From there, teams can use paths and filters to route, block, or escalate. Current capabilities include the following:

• PII Detection -- Scans AI-generated text for more than 30 types of personally identifiable information, including Social Security numbers, credit card numbers, bank details, email addresses, and physical addresses. Detected PII can be automatically blocked or redacted before it moves downstream.
• Prompt Injection Blocking -- Reviews user or external input before it reaches an AI model, catching attempts to manipulate the model's behavior.
• Jailbreak Detection -- Flags attempts to bypass an AI model's built-in safety controls.
• Toxicity Detection -- Screens content for hate speech, threats, insults, and other harmful language before it gets published, forwarded, or stored.
• Sentiment Analysis -- Gauges the tone of AI-generated or user-submitted content with confidence scores, so teams can route negative or mixed-sentiment outputs for human review.

AI Guardrails works across Zapier's platform. In Zaps, teams add a guardrail step after any AI action. In Agents, it functions as a tool the Agent is instructed to use before acting on AI output. And through MCP, AI clients like Cursor and Claude can call guardrail actions directly.

"The conversation around AI safety usually stops at 'we wrote a policy,'" Sammut said. "What teams actually need is something that runs in the background and catches problems before they become incidents. That's what this does."