AegisAI Security

AI agents are moving into email, workflows and production systems faster than security teams can write policy for them. On Thursday, October 29, in New York, we’re hosting a small dinner on how to secure them, with Shan Rao, founder of Google’s Secure AI Framework (SAIF). Shan builds security for Google’s AI models and agents, across Gemini, Cloud, Search, Workspace, Chrome and Android. We’ll get into three things: ➡️ How Google threat models AI agents ➡️ Where agentic systems break: prompt injection, over-permissioned tools, and data leaking through the agent ➡️ What a practical enterprise security framework for AI looks like It’s invitation only and off the record, for CISOs and security leaders. Seats are limited! Request to join here: https://luma.com/35os97ld

We’re going live for Cybersecurity Awareness Month. Attackers now use AI to write phishing emails that read like they came from your CFO, your vendor, or your IT team. Legacy filters weren’t built for that. On Tuesday, October 27, AegisAI Security CEO and Co-Founder Cy Khormaee and Aaron Daily of BorderLAN Security will talk through: → How AI is changing phishing and BEC attacks → Why traditional email defenses are falling behind → Practical steps security teams can take this quarter Stick around for live Q&A at the end. 📅 Tues, Oct 27 | 12 PM ET | LinkedIn Live Click “Attend” below to save your spot 👇 #CybersecurityAwarenessMonth #EmailSecurity #Phishing #CISO #AISecurity

PHP2Shell is a website vulnerability. Its biggest impact may land in your inbox. Disclosed this week, PHP2Shell lets attackers take over unpatched sites that run PHP, including WordPress, Drupal and Joomla. It was found with the help of AI, which also made it easier to build working attacks. Attackers have access to the same tools. Every hacked site comes with something email security relies on: a domain with years of clean history. Here is what we expect next: Links in legitimate emails lead to phishing pages once the site behind them is hacked. Phishing moves onto established domains that pass age and reputation checks. Hacked websites become hacked senders, with stolen mail credentials that pass SPF, DKIM and DMARC. Attackers were already heading this way. In our Q3 State of the AI Attack Report, the share of mass AI phishing sent from compromised accounts rose 38x in one quarter, from 0.2% to 7.7%. When AI finds the flaws and writes the lures, a trusted sender is no longer proof of a safe message. You can only fight AI with AI. See our State of the AI Attack Report for related findings: https://lnkd.in/gR25uWSM #EmailSecurity #Phishing #PHP2Shell #AI #Cybersecurity

Unlock your security potential! Book a demo with AegisAI Security and receive a FREE pair of AirPods. Experience cutting-edge protection while enjoying your favorite tunes.

Today we are incredibly proud to announce Proteus - our first foundational model designed specifically to secure the inbox. Ryan Luo and I spent years protecting billions of users at Google across systems like #safebrowsing and #reCAPTCHA. When we saw this AI shift accelerating, we knew that simply wrapping a general-purpose frontier model wasn't going to be enough. We needed a model trained from the ground up for this specific threat. Instead of relying on rigid rules or assumptions about user behavior, Proteus actually investigates every single message. It cross-checks the claims an email makes against external evidence the sender does not control - before it ever reaches an employee. We put Proteus up against a benchmark of over 3,000 human-reviewed phishing attacks drawn from live production traffic. The results speak for themselves: 🛡️ >99% of attacks identified. 📉 ~30x fewer missed attacks compared to leading general-purpose open and closed-source models. ⚡️ 7x faster median verdicts (and up to 26x faster on the hardest edge cases). For most of the history of cybersecurity, defenders had one fundamental advantage: novelty was expensive. Crafting a flawless, personalized spear-phishing campaign took serious human effort, meaning attackers had to reuse their work to make a profit. That reuse was our best weapon - we could observe an attack once, characterize it, and block it everywhere. Today, an AI-automated spear-phishing email costs almost nothing to produce, draws the same 54% click-through rate as one written by a human expert, and every single attack is a first occurrence. In our own observations at AegisAI, AI-generated spear phishing has jumped from less than 1% to over 20%+ of total phishing volume this year. When your adversary is generating novel attacks at machine speed, writing detection rules at human speed means you've already lost. And relying on behavioral anomaly detection doesn't work when a hijacked mailbox writes from inside its own trusted baseline. You cannot patch human trust. When the attack is AI, the defense has to be AI. Our team at AegisAI Lab Bin Lu Pavel Petrenko Yogesh Patil Sam M. have put together a deep dive into how Proteus works, why legacy anomaly detection is structurally failing against AI threats, and the full data behind our benchmarks. I've always believed the goal of security isn't some unattainable perfection; it's to make the attacker's business model unprofitable so they go do something else. Proteus shifts the economics back in favor of the defender. Check out the full announcement here: https://lnkd.in/gTRXH_3q #Cybersecurity #GenerativeAI #EmailSecurity

The best security conversations rarely happen on a conference floor. Join a small group of CISOs for an evening of great food and frank peer exchange at Bull & Bear Orlando.