Founded by former Google Safe Browsing and reCAPTCHA engineers, AegisAI launched with $13M in seed funding and a blunt argument: the tools Proofpoint and Mimecast built to stop human-crafted attacks aren't built for the ones machines now generate at scale.
The numbers are hard to argue with. AegisAI's own research, presented at the Messaging, Malware and Mobile Anti-Abuse Working Group conference earlier this year, found that AI-generated email attacks grew five times over in 2025. Sophisticated threats are now bypassing traditional security filters more than half the time. That's not a marginal shift in the threat terrain. It's a structural one.
Cy Khormaee and Ryan Luo know that terrain as well as anyone. Luo spent nearly a decade at Google leading engineering for reCAPTCHA and Safe Browsing, the systems that screen billions of users from phishing attempts daily. Khormaee ran product for those same platforms. When they left to start AegisAI in 2025, their pitch to Accel and Foundation Capital, who co-led the $13M seed round announced in September 2025, was essentially this: the attackers have already adopted AI, and the defenders haven't.
Their platform deploys autonomous multi-agent AI to inspect every component of an incoming email: links, attachments, metadata, QR codes. It's designed to intercept spear phishing, business email compromise, and malware before messages reach inboxes. That matters. Compared with legacy tools, AegisAI claims a 90% reduction in false positives, which, frankly, is the metric that tends to break enterprise security teams. A system that cries wolf constantly gets tuned down. The real threats slip through in the noise.
The incumbents and the numbers
Proofpoint and Mimecast dominate a market that Mordor Intelligence pegged at roughly $7.25 billion in 2026, growing toward $12 billion by 2031. Both companies built their platforms around secure email gateways, inline inspection tools that require an MX record change to deploy and were architected for a world where attackers were human, methodical, and working from a limited playbook. That world is gone.
KnowBe4's 2025 Phishing Threat Trends Report found that 82.6% of phishing emails analyzed between September 2024 and February 2025 contained AI-generated content. Mimecast's own 2025 Global Threat Intelligence Report counted 9.13 billion threats in the first nine months of the year, with phishing accounting for 77% of them. ClickFix social-engineering chains, a tactic that tricks users into executing malicious commands themselves, grew 500%. These aren't exploits the old guard designed around.
AegisAI's March 2026 Vanguard launch, demonstrated live at RSA Conference, pushed the company further into territory Proofpoint and Mimecast have been slow to occupy. Vanguard is an autonomous agent built specifically to defeat adversarial CAPTCHAs, the evasion tactic attackers use to hide malicious payloads behind interactive challenges that legacy scanners can't solve. The team that built Google's reCAPTCHA is now building the thing that beats it when criminals turn it against defenders. That's a pointed credential.
The incumbent response, at least from Mimecast, has been an API-compatible deployment push that lets customers avoid MX record changes, a rollout designed to reduce friction, not rethink the architecture. Proofpoint's NexusAI platform adds machine learning on top of a gateway model that's still fundamentally sequential and rule-adjacent. Neither approach looks built for a world where every attacker has access to generative models that can write a convincing spear phishing email in your CFO's style in seconds.
What changes when the cost of attack collapses
Here's the real issue: the commodity shift in attack tooling changes the economics permanently. Spear phishing used to require hours of research per target. Now it's a prompt and a few seconds. That changes the volume, the personalization, and the success rate all at once, and it does so in a way that any halfway decent LLM can replicate. Defenders who rely on signature matching, reputation databases, and heuristic rules built against historical attack patterns are bringing last year's answer to a different question.
AegisAI is entering early customer testing for Vanguard later this year. The $13M seed was enough to build the product and prove the architecture. Whether the company can scale past the pilot stage, and whether enterprise security buyers will move fast enough to matter, are the only questions left worth asking. The threat already moved.
Also read: Runway bets on being the infrastructure layer under every generative media model • Meta and OpenAI commit 12 gigawatts to AMD's new AI chips as Nvidia's grip on data centers starts to slip • Microsoft is replacing OpenAI's image models in PowerPoint and Bing with its own MAI-Image-2
