August Monthly CASSUG Meeting

Greetings, data enthusiasts!

Our August meeting is scheduled for Monday, August 17th, at 5:30 pm! We will meet in person at: Troy Web Consulting, 135 Mohawk St, Cohoes, NY 12047.

Please RSVP if you are attending, so we can purchase an appropriate amount of food for everyone. Food will be a variety of Korean BBQ and sides from K-Plate in Troy. For this meeting, we are having a session from Ben Harris discussing multimodal agents.

Our meeting schedule is as follows:

5:30 PM: Food, soft drinks, and networking
6:15 PM: Chapter news and announcements
6:30 PM: Presentation

We usually wrap up between 7:30 PM and 8:00 PM. Here are the details for our session:

Ben Harris: Can Multimodal Agents Build the Next Signal? Exploring the Limits of AI in High-Stakes Privacy Engineering

Everybody wants to know if autonomous AI agents are finally ready to write production-grade software. But what happens when we point a natively multimodal AI system at the ultimate engineering stress test: designing and implementing a post-quantum, Tor-routed privacy messenger aiming for the security bar set by apps like Signal?

Using Knot—an experimental Rust-first privacy messenger featuring a native Android application, slated for future public release at https://github.com/DolosLabs/Knot—as our conceptual benchmark, we’ll explore the theoretical capabilities and limits of AI in high-stakes privacy engineering. Because modern AI models are designed from the ground up to reason seamlessly across text, images, video, audio, and code, the scope of what an agent can build has fundamentally expanded. But cryptography and secure communications still don't do "almost right."

Instead of making definitive claims, we’ll investigate the architecture to see where agents might naturally excel and where they are likely to be tested:

• The Blueprint: Could an AI successfully combine post-quantum algorithms (like ML-KEM-1024) with classical tools (like X25519 and OPAQUE) to architect defenses against "harvest-now-decrypt-later" attacks?
• The Plumbing: Is it possible for an agent to correctly manage encrypted local storage (via SQLCipher and Android Keystore) alongside a complex hybrid Double Ratchet without breaking state across Tor streams?
• The Nuance: Can natively multimodal agents parse high-level protocol logic and design constraints—such as understanding simple UI mockups and transforming them into working Jetpack Compose code, implementing strict resource limits to prevent infinite loops, and grasping the subtle differences between message-level deniability and participation deniability?

Ultimately, this talk is an exploration of the unknown. We’ll map out the massive gap between generating a theoretical codebase and shipping a secure, deployable APK, and discuss whether multimodal AI agents have what it takes to eventually cross that divide.