Create Email Guardrails for AI Agents

Material finds and secures sensitive information in inboxes, preventing AI agents from accidentally leaking confidential information or accessing password resets and magic links.

Problem: AI agents can see and act on anything in the inbox

Busy execs, sales reps, and customer support team members use AI tools to help them manage and scale communications with a huge variety of external stakeholders. An AI agent can act like a personal assistant for email, but tools like Gemini, Superhuman, and Claude Cowork are doing much more than just writing drafts and checking spelling. Today’s AI agents can read, search, and take actions in the inbox. While the time savings are real, so is the new risk introduced by agents misinterpreting prompts or exposing sensitive information.

• Lack of visibility and control: Most security teams cannot definitively answer one simple question: "What sensitive data is stored in inboxes?" Without this visibility, it’s impossible to predict what an agent will do when (not if) it accesses this information.
• Sensitive data leaks: When an agent drafts an email, it can pull in any data it scrapes from the contents of the inbox. If sensitive data like banking info, PII, and confidential customer info is accessible to an agent, any of this information can be incorporated into outbound emails. A busy human operator may miss this and send data they shouldn’t.
• Password resets escape the inbox: The inbox isn’t just a communication platform, it’s also a gateway into other applications. A helpful agent may spot a magic link or a password reset code and forward it where it doesn’t belong.

Solution: Restrict agent access to sensitive email

Material Security identifies and secures sensitive email, prompting human users with a step-up challenge when they need access. An AI agent can’t act on what it can’t read, so sensitive info stays put.

• Protect sensitive messages with MFA: Material automatically finds messages containing sensitive content (like financial data, PII, or legal documents), password resets, and magic links and redacts them. To view the original message, the human user must complete a real-time MFA prompt. Meanwhile, that information is out of AI-sight, out of AI-mind for agents.
• Guardrails for agents, no prompt engineering needed: Protecting email at the source means employees don’t have to be advanced AI users to set guardrails for their agents. Out of the box, Material knows what sensitive content to look for and lock down. 
• Seamless user experience: It’s simple for an authorized human operator to access a protected message. Just a quick, familiar MFA flow through an approved app opens the door to what an employee needs.