Artificial intelligence email security startup StrongestLayer Inc. today unveiled two innovative products designed to enhance email security, specifically for transactions involving money or interactions with AI assistants.
The first product, Intent Workflows, is engineered to place holds on payment, payroll, and executive account requests pending verification before any action is taken. Complementing this is AI Instruction Intercept, which is specifically designed to prevent hidden instructions within a message from being read by AI assistants such as Microsoft Corp.’s Microsoft 365 Copilot.
Intent Workflows strategically targets attacks that do not utilize malware or send suspicious links. In instances of vendor email compromise, a criminal may seize control of a legitimate supplier’s mailbox and send altered bank details within an ongoing email conversation.
During the second quarter, StrongestLayer analyzed 32,052 attacks that bypassed existing defenses across 45 organizations. Notably, only 3.6% of these attacks constituted payment or bank change requests. However, this small percentage accounted for over 98% of the estimated financial exposure under a loss model based on the Factor Analysis of Information Risk framework. Sender authentication measures proved largely ineffective, as 94% of these requests successfully passed through those checks.
According to Alan LeFort, StrongestLayer’s co-founder and CEO, traditional email security primarily evaluates whether a message appears suspicious. He stated, “The costliest attacks are built to look fine.” StrongestLayer’s approach focuses on understanding the intent behind a message, thereby ensuring that for actions like wire transfers, the verification process is prioritized over presumptive judgment.
StrongestLayer’s detection mechanisms continue to monitor each email message. Intent Workflows introduces a mechanism to hold specific requests that could not be reversed, allowing each customer the discretion to determine whether these requests await verification, are routed to the security team, or directed to a designated process owner.
Organizations also have the option to allow a request to proceed with a policy banner that appears solely on tracked requests, communicating companies’ rules, such as “Bank change: call the number on file.” All payments subjected to scrutiny are placed on hold regardless of the final verdict, preventing a compromised vendor account from processing illegitimate transactions.
The software evaluates each payment request against the payee’s historical data, with customers setting the threshold for scrutiny—be it every payment or only those exceeding $5,000. If no discrepancies are detected, the payment owner can confirm with a single click, either via Microsoft Teams or email. Any alterations, or if a new vendor is introduced, necessitates a follow-up call to a pre-existing number, and the email will not be released until the outcome is duly logged.
Payroll adjustments are verified through the same protocol, while requests involving executive passwords, multi-factor authentication, and account recovery resets are initially directed to the identity or security teams, even when they originate from the help desk.
Eric Sanchez, Chief Information Security Officer at Orrick, Herrington & Sutcliffe LLP, remarked that expecting employees to remain vigilant with every email has historically proven ineffective. As a customer and design partner of StrongestLayer, Sanchez noted that under the new protocol, every irreversible payment now undergoes “a deliberate check by a named owner.”
The announcement also addresses emerging threats. Last year, researchers at Aim Security Ltd. identified the EchoLeak vulnerability in Microsoft 365 Copilot, which allowed crafted emails to cause the assistant to leak sensitive data.
AI Instruction Intercept evaluates each email prior to delivery. A reasoning model screens it for hidden instructions in alignment with the customer’s business context. Only after this analysis is the email forwarded to AI systems such as Copilot, Google LLC’s Gemini, or a security classifier. By default, messages deemed suspicious are quarantined with a clear explanation of the concealed instructions detected.
Email represents a potential access point to AI assistants, one that “any outsider can utilize,” said Joshua Bass, co-founder and Chief Product Officer of StrongestLayer. He emphasized that securing the inbox is the first line of defense against prompt injection threats, as attackers require only an email address to initiate their efforts.
Both products are now available for Microsoft 365 and Google Workspace, operating in conjunction with existing email security solutions.