The email sat in LangChain's inbox looking perfectly legitimate. No typos, no broken English, no obvious red flags that might have tripped traditional security filters. That's because—unlike the clumsy phishing attempts of years past—this one had been written by artificial intelligence.
It's a scenario playing out thousands of times across corporate America, according to AegisAI, a San Francisco email security startup that just closed a $36 million Series A round. Battery Ventures led the July funding, with Accel and Foundation Capital returning as investors. Total capital raised since the company's 2025 founding: $49 million.
The timing feels almost too on-the-nose. Research that AegisAI released tracked a fivefold surge in AI-generated spear phishing during 2025. The company's analysis—covering more than 20,000 malicious emails—found that AI-crafted attacks jumped from 2.8% to 13.9% of all phishing attempts over a single year. Even more unsettling: according to AegisAI's data, these messages reached inboxes more than half the time. And 72.6% of the successful AI attacks sailed right through authentication protocols designed to catch them.
"Pattern matching doesn't work anymore," says Cy Khormaee, AegisAI's CEO and a former Google security engineer. The old playbook—scanning for known malicious signatures, flagging suspicious keywords—assumes attackers make mistakes. AI doesn't make those mistakes.
The Google Connection
Khormaee and his co-founder Ryan Luo, who serves as CTO, both spent time building Google's security infrastructure: Safe Browsing, reCAPTCHA, Web Risk. That pedigree matters in a field where trust is the product. Their pitch revolves around what they call a fundamental rethinking—deploying autonomous AI agents to analyze the intent and identity behind each message, rather than just its content.
According to AegisAI's internal testing, this approach cuts false positives by as much as 90% compared to legacy solutions. (That figure hasn't been independently verified, and the company acknowledged as much.)
Early customers include LangChain, the AI developer tools company; Mesh, a crypto payments platform; and Lokker, which handles privacy compliance. A case study from March detailed how LangChain used the platform to block what it described as thousands of AI phishing attacks—though the company didn't specify the timeframe or provide granular metrics.
The product itself has evolved quickly. There's the core inbound defense system, naturally. Then there's Vanguard, an agent introduced in March that "hunts threats beyond the inbox"—the company's language, suggesting some proactive capability beyond passive filtering. And just before this funding round closed, AegisAI rolled out its Red Team Agent: an internal simulator that launches AI-driven spear-phishing attacks against approved employee groups. Think of it as a fire drill, except the fire is algorithmic.
The $20 Billion Problem
Context matters here. The FBI's Internet Crime Complaint Center reported $20.88 billion in total cybercrime losses for 2025, with business email compromise alone accounting for roughly $11.37 billion of that staggering figure. An earlier FBI advisory from 2024 pegged BEC as a $55 billion cumulative scam when international losses were factored in.
Those numbers—sprawling, almost abstract—help explain why investors keep writing checks to security startups. Battery Ventures' Dharmesh Thakker led this round; AegisAI's $13 million seed, announced in September 2025, was co-led by Accel's Eric Wolford and Foundation Capital's Sid Trivedi.
The new capital will fund what the company calls its "fleet of autonomous defense agents," push Vanguard toward general availability, and expand enterprise sales. LinkedIn data from mid-2026 listed AegisAI's headcount between 11 and 50 employees. The company recently brought on Jeremy Riesenfeld as VP of Sales—a signal that go-to-market is the next frontier.
An Arms Race in Code
Of course, AegisAI isn't alone in this space. Proofpoint and Mimecast have long dominated enterprise email security. Newer players like Abnormal Security and Material Security have raised substantial rounds of their own, each promising some variation on AI-powered threat detection.
Which raises the central question: Can AI-native defenses actually keep pace with AI-native attacks? Or are we entering a cycle where each incremental improvement in phishing sophistication will be met with an equally incremental defensive response—an arms race measured in basis points and false positive rates?
Perhaps more pressingly, will enterprises bet their inboxes on autonomous agents? Email remains the front door to most corporate networks. Handing over the keys to an algorithm, even a sophisticated one, requires a leap of faith that some security teams may not be ready to make.
For now, AegisAI seems to be betting that the answer is yes. The phishers have gone algorithmic. The question is whether the rest of us will follow.
