AegisAI Launched Proteus: AI vs AI in Your Inbox (and How to Read Its 99% Claim) | inicIA
Direct answer
On October 8, 2026, startup AegisAI announced Proteus: an artificial intelligence model built specifically for email security. Instead of filtering with static rules, Proteus investigates each message —sender identity, embedded links, requested action— before it reaches your inbox. The company claims it detected more than 99% of 3,020 human-reviewed phishing attacks in its internal benchmark, versus 81% for an unnamed leading closed-source general-purpose model and 80% for an unnamed open-weight model. That figure is self-reported and has not been independently audited. Behind the announcement is a serious thesis: AI-crafted attacks now bypass traditional controls at nearly twice the rate of human-written ones, so the defense has to be AI too. AegisAI has raised $49 million ($36M Series A on July 23, 2026 led by Battery Ventures, after a $13M seed in Sep 2025 with Accel and Foundation Capital).
Why this matters for your business
- The attack changed its nature: AegisAI's study presented at the 2026 M3AAWG conference, covering more than 20,000 phishing and scam emails, found that AI-generated spear phishing grew from 2.8% to 13.9% of all observed phishing in a single year.
- Email is the front line: the FBI's 2025 Internet Crime Report recorded record losses of $20.8 billion, with business email compromise (BEC) accounting for $11.64 billion —versus under $52 million for ransomware and malware combined.
- The tactic is personalization: attackers use language models to research your employee, map their professional relationships, and write a tailored lure. "They've researched you, they understand everything about you," CEO Cy Khormaee summarized.
What Proteus promises to do differently
- Analyze intent, not signatures: instead of matching known patterns, the system evaluates the intent behind a message —unusual sender behavior, subtle language deviations, metadata that doesn't add up.
- Browse like a human: its companion agent, Vanguard (announced in March), follows suspicious links and attachments onto the open web and returns a threat report within minutes.
- Integrate without friction: it connects to Microsoft 365 and Google Workspace via API, with no MX record changes required.
The honest part: how to read the "99%"
The figure compares the specialized model against two unnamed general-purpose models —a comparison favorable by design. Also, AegisAI disclosed no pricing or customer count for Proteus (at the company's 2025 public launch it reported three paying customers: Lokker, Mesh Connect and others). None of this invalidates the product; it simply sets how to evaluate any AI security promise:
- Ask for independent benchmarks: self-reported figures are the starting point, not the proof.
- Ask about your false positives: 99% detection is useless if it blocks legitimate customer emails.
- Measure against your real threat: if your team receives generic lures, a traditional filter is enough; if they move money by email, you need agentic defense.
- The human link remains: no tool eliminates the need for verification protocols on transfers and sensitive requests.
Related reading
- The CEO deepfake on a video call that cost $25.6M: the protocol that would have prevented it
- Privacy and data: what happens to your customers' information
- 3 real businesses that took off with AI in 2026 (with verified numbers)
Frequently asked questions
What is AegisAI's Proteus?
An AI model announced on Oct 8, 2026, built specifically for email security: it investigates sender identity, links and the requested action of each message before it reaches the inbox, instead of using static rules.
Is the 99% detection figure real?
It is the figure AegisAI reports from its own benchmark (over 99% on 3,020 human-reviewed phishing attacks). It is self-reported and has not been independently audited; take it as the starting point, not as proof.
Who is behind AegisAI?
Cy Khormaee (CEO) and Ryan Luo, both former Google security leads who worked on reCAPTCHA, Safe Browsing and Web Risk. Founded in 2025, it has raised $49M with customers including Mesh, LangChain and Lokker.
Does my business need AI-powered email defense?
It depends on your attack surface: if your team authorizes payments or shares sensitive data by email, personalized AI phishing is your primary threat and deserves a matching defense. Start with human verification protocols, which cost nothing.
Take the next step
AI-crafted phishing can no longer be spotted by eye: it takes systems to detect it. If your team handles money or sensitive data by email, talk to inicIA on WhatsApp: https://wa.me/17866120930
Chat on WhatsAppBy inicIA
