Controlling AI Models from the Inside
Episode
43 min
Read time
2 min
Topics
Startups, Fundraising & VC, Artificial Intelligence
AI-Generated Summary
Key Takeaways
- ✓Internal Model Instrumentation: Rynx analyzes which subregions of AI models activate during generation, identifying when prohibited content emerges before completion. This catches jailbreaks that bypass traditional prompt and response filters, similar to monitoring hallways inside a building rather than just checking IDs at the entrance gate.
- ✓Cost Reduction at Scale: Traditional guardrails require running an 8 billion parameter guard model twice (prompt and response), totaling 160 billion parameters of inference. Rynx reduces this to 20 million parameters—a thousand-fold improvement—making safety economically viable for video, audio, and edge device deployments where current solutions are too expensive.
- ✓Context-Specific Safety Customization: Every industry requires different safety policies beyond general prohibited content. Law firms need different protections than medical shops or customer service applications. Rynx builds customizable safety modules that work with any off-the-shelf model without requiring retraining or fine-tuning the primary model.
- ✓Edge Device Viability: Current guardrail approaches cannot deploy on edge devices because memory constraints barely accommodate the primary model. With only 20 million parameters overhead, Rynx enables comprehensive safety on resource-constrained devices where quantization already pushes hardware limits and no room exists for separate guard models.
- ✓Defense-in-Depth Architecture: Effective AI safety requires multiple layers working together—prompt filters, response analyzers, system-level features, and model-internal monitoring. Combining Rynx's model-native safety with traditional guardrails creates robust protection, similar to how national security requires military, border control, and local law enforcement operating simultaneously.
What It Covers
Ali Khatri, founder of Rynx, explains how traditional AI guardrails only filter inputs and outputs while his company instruments model internals to detect unsafe behavior during generation. This approach delivers comparable safety performance at 1/1000th the computational cost by analyzing internal model states rather than running separate guard models.
Key Questions Answered
- •Internal Model Instrumentation: Rynx analyzes which subregions of AI models activate during generation, identifying when prohibited content emerges before completion. This catches jailbreaks that bypass traditional prompt and response filters, similar to monitoring hallways inside a building rather than just checking IDs at the entrance gate.
- •Cost Reduction at Scale: Traditional guardrails require running an 8 billion parameter guard model twice (prompt and response), totaling 160 billion parameters of inference. Rynx reduces this to 20 million parameters—a thousand-fold improvement—making safety economically viable for video, audio, and edge device deployments where current solutions are too expensive.
- •Context-Specific Safety Customization: Every industry requires different safety policies beyond general prohibited content. Law firms need different protections than medical shops or customer service applications. Rynx builds customizable safety modules that work with any off-the-shelf model without requiring retraining or fine-tuning the primary model.
- •Edge Device Viability: Current guardrail approaches cannot deploy on edge devices because memory constraints barely accommodate the primary model. With only 20 million parameters overhead, Rynx enables comprehensive safety on resource-constrained devices where quantization already pushes hardware limits and no room exists for separate guard models.
- •Defense-in-Depth Architecture: Effective AI safety requires multiple layers working together—prompt filters, response analyzers, system-level features, and model-internal monitoring. Combining Rynx's model-native safety with traditional guardrails creates robust protection, similar to how national security requires military, border control, and local law enforcement operating simultaneously.
Notable Moment
Khatri reveals that publicly available audio, video, and image generation models from major companies generate prohibited content with minimal effort because the economics of traditional safety make comprehensive protection impossible. Companies ship unsafe models rather than double their inference costs, creating widespread vulnerabilities that average users can exploit without sophisticated techniques.
Episode Transcript
Welcome to the Practical AI podcast, where we break down the real world applications of artificial intelligence and how it's shaping the way we live, work, and create. Our goal is to help make AI technology practical, productive, and accessible to everyone. Whether you're a developer, business leader, or just curious about the tech behind the buzz, you're in the right place. Be sure to connect with us on LinkedIn, x, or Blue Sky to stay up to date with episode drops, behind the scenes content, and AI insights. You can learn more at practicalai.fm. Now onto the show. Welcome to another episode of the Practical AI Pod This is Daniel Whitenak. I am CEO at Prediction Guard, and I'm joined as always by my cohost, Chris Benson, who is a principal AI research engineer at Lockheed Martin. How are you doing, Chris? Hey. Doing great today, Daniel. How's it going? It's going well. Done a a little bit of snow shoveling on the ground. As we speak, we're kind of headed into, winter break or or the holiday Christmas season here in the in The US. And I think this this episode will be released in the new year. So if you're listening to this, this you're listening in the future. To be honest, I'm really excited about talking about the future because our guest today is really thinking very innovatively about how we can secure our AI models and and have safety as we move into that future. Really excited to welcome to the show today, Ali Khatri, who is founder of Rynx. Welcome, Ali. Thanks. Thanks for having me, Daniel. Yeah. Yeah. It's, we met earlier this fall. Really fascinated by your kind of line of work and and technological innovation at at Rynx. But I I also know that you've been thinking about these topics around AI safety, guardrails, disallowed content, etcetera, for for quite some time. Could you give us a little bit of a background of how you got into these topics and and what you've done in the past? Yeah. So I have been in the machine learning for AI safety or anti abuse use cases in general, for the past eight or so years of my career. I spent about three years at Meta where I built infrastructure that serves about half of the world's population. Basically, any safety anytime you type a message on Facebook, it goes through tens of safety checks which are powered by thousands of models. I built the infrared these models run on. And then I moved on to Roblox where I built AI powered like, where I built, systems to protect about $3,000,000,000 in payments against fraud. So I've been in this space. I've been using AI models to sort of protect against abuse. And during this time, I realized that the models that I'm using themselves are susceptible to abuse. So that's what, led me to family breaks. And I know that now you're thinking …
Get the full transcript (7,530 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
You just read a 3-minute summary of a 40-minute episode.
Get Practical AI summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
More from Practical AI
Less about Models; More about Architecture
Sep 3 · 45 min
David Senra
Lulu Cheng Meservey, Founder of Rostra
Aug 12
More from Practical AI
Building the Foundation for the Agentic AI Era
Aug 28 · 45 min
20VC (20 Minute VC)
20VC: Leo Aschenbrenner's Largest Holding: Inside the $90BN Bloom Energy | Why Electricity, Not AI Models, Will Decide the Winners of the AI Race | Why We Are Not in an AI Capex Bubble | Energy Sovereignty and The Future of Power with KR Sridhar
Jun 29
Books, tools, and gear mentioned in this episode
SignalCast may earn commission on purchases via these links.
Tools
“Sponsors: Miro”
“Sponsors: Prediction Guard”
company
- RynxBy guest
“Ali Khatri, founder of Rynx, explains how traditional AI guardrails only filter inputs and outputs while his company instruments model internals to detect unsafe behavior during generation.”
More from Practical AI
We summarize every new episode. Want them in your inbox?
Less about Models; More about Architecture
Building the Foundation for the Agentic AI Era
AI Proficiency: From Users to Builders
Models, Harnesses, and Multi-Agent Systems
Reconstructing how OpenAI agents attacked Hugging Face
Similar Episodes
Related episodes from other podcasts
David Senra
Aug 12
Lulu Cheng Meservey, Founder of Rostra
20VC (20 Minute VC)
Jun 29
20VC: Leo Aschenbrenner's Largest Holding: Inside the $90BN Bloom Energy | Why Electricity, Not AI Models, Will Decide the Winners of the AI Race | Why We Are Not in an AI Capex Bubble | Energy Sovereignty and The Future of Power with KR Sridhar
Eye on AI
May 6
Loris Degioanni: Why AI Is Breaking Cybersecurity, and What Comes Next
Software Engineering Daily
Apr 21
Unlocking the Data Layer for Agentic AI with Simba Khadder
Masters in Business
Feb 26
At The Money: Diversifying with Managed Futures ETFs
Explore Related Topics
This podcast is featured in Best AI Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's Startups & Product Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into Practical AI.
Every Monday, we deliver AI summaries of the latest episodes from Practical AI and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime