Building a Social Engineering Defense Platform to Protect Both Humans and AI Agents

With CISOs prioritizing agent security, one question I've heard is, "why should I prioritize social engineering defense when it has traditionally focused on the human layer?" The answer is the next wave of social-engineering attacks does not care who is on the other side, whether they are human or an AI agent.

As agents take over more and more human workflows and social interactions, Social Engineering Defense (SED) has to protect both humans and agents. We're already seeing this today in how Doppel's customers use our core products - Human Risk Management (HRM), Digital Risk Protection (DRP), and Email Security.

1) Recently we have observed our social engineering red teaming agents call into customer support lines powered by AI agents, leading to agent vs. agent battles on the phone. Doppel’s HRM today already provides continuous pen testing of both humans and agents across channels at scale. As we expand from human to also non-human risk management, we’re rethinking security awareness training for agents and risk management for the exponentially growing number of identities operating in the enterprise IT stack.

2) Lately we have seen GEO / AEO poisoning attacks distribute cred theft links via LLM surfaces. Doppel’s DRP today already takes down the impersonations, fake personas, and PII leaks that spread across the digital landscape, whether traditionally crafted or AI generated. As AI continues to take over digital content, we’re expanding DRP to cover every surface in which LLMs generate and consume information to prevent agent-initiated fraud or credential theft as well as non-human insider threats.

3) In just the past few weeks, we are seeing AI personal assistant agents like Muse and Instinct go viral. These agents move fast and consume emails before traditional security solutions. Doppel’s email security today covers both human and agent-based threat vectors like prompt injection. In parallel, we’re building architectures that can protect these low-latency AI personal assistant agents and guard against agentic DDOS attacks.

To compromise an organization, social engineering continues to be the most profitable way in. Humans used to be the “most vulnerable link” – now it’s also agents that need to be protected. Whether the interaction is human-to-human, human-to-agent, or agent-to-agent, the objective remains the same: manipulate trust to drive an action. This is why Doppel is building the platform to defend the entire social engineering attack chain, for both humans and agents.