Subvolts take AI's autonomous agents demand new security thinking, moving beyond traditional defenses. Oct 21, 2025 - Developers
- Learn about the new security challenges posed by autonomous AI agents and how to build more robust systems.
- Business
- Understand how AI's evolving capabilities impact enterprise security and national security layers.
This interview provides insights into the evolving field of AI security from a company working directly with major AI labs.
Securing the AI Frontier: Irregular Co-founder Dan Lahav
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Sequoia Capital's Training Data podcast features Dan Lahav, co-founder of Irregular, discussing "frontier AI security." Lahav explains how autonomous AI models acting as independent agents create new cybersecurity challenges. He details how emergent AI behaviors, like social engineering and evading defenses, necessitate a fundamental rethinking of security strategies for governments and enterprises.
Imagine AI models acting like independent agents. This episode explains how they might trick each other or bypass computer defenses. It's important because we need new ways to protect our systems as AI gets smarter and more independent.
This episode explores the cutting edge of AI security, a critical area as frontier labs like OpenAI, Anthropic, and Google DeepMind develop increasingly autonomous AI. Listener learns about the shift from traditional defenses to proactive strategies against AI-on-AI threats. It is for security professionals, AI developers, and policymakers concerned with national security implications. The discussion directly relates to current efforts by major AI labs to understand and control advanced AI capabilities.
- HostSequoia Capital
- GuestDan Lahavco-founder, Irregular
- Guest: Dan Lahav, co-founder of Irregular
- Show: Training Data (Sequoia Capital)
- Length: 44 minutes
- Focus: Frontier AI security and autonomous AI agents
- Collaborators: OpenAI, Anthropic, Google DeepMind
Irregular co-founder Dan Lahav is redefining what cybersecurity means in the age of autonomous AI. Working closely with OpenAI, Anthropic, and Google DeepMind, Dan, co-founder Omer Nevo and team are pioneering “frontier AI security”—a proactive approach to safeguarding systems where AI models act as independent agents.
Dan shares how emergent behaviors, from models socially engineering each other to outmaneuvering real-world defenses like Windows Defender, signal a coming paradigm shift. Dan explains why tomorrow’s threats will come from AI-on-AI interactions, why anomaly detection will soon break down, and how governments and enterprises alike must rethink defenses from first principles as AI becomes a national security layer.
Hosted by: Sonya Huang and Dean Meyer, Sequoia Capital
00:00 Introduction 03:07 The Future of AI Security 03:55 Thought Experiment: Security in the Age of GPT-10 05:23 Economic Shifts and AI Interaction 07:13 Security in the Autonomous Age 08:50 AI Model Capabilities and Cybersecurity 11:08 Real-World AI Security Simulations 12:31 Working with AI Labs 32:34 Enterprise AI Security Strategies 40:03 Governmental AI Security Considerations 43:41 Final Thoughts
- emergent behaviors
- Dan Lahav
- unexpected capabilities or actions that arise in AI models as they become more complex.
- social engineering
- Dan Lahav
- manipulating people into performing actions or divulging confidential information, applied here to AI models interacting.
- anomaly detection
- Dan Lahav
- identifying unusual patterns or outliers in data, which may become less effective for AI security.
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
1 h 46The OpenAI Internet Browser Has Arrived: ChatGPT Atlas w/ Dave Blundin & Alexander Wissner-Gross | EP #203
WithDave Blundin, Alexander Wissner-Gross
Questions people ask
- What is frontier AI security?
- It is a proactive approach to safeguarding systems where AI models act as independent agents, pioneered by Irregular.
- What kind of threats does Dan Lahav discuss?
- He discusses threats from AI-on-AI interactions, emergent behaviors like social engineering, and AI models outmaneuvering real-world defenses.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.