← Multimodal Sources 🕐 2 min read
Multimodal Sources

Securing the AI Frontier: Irregular Co-founder Dan Lahav

> Models are able to do that. We were able to show how a model was able to take out Windows Defender, the real Windows Defender, in a Tor environment, granted, but the real Windows Defender.

Show: Training Data · Publisher: Sequoia Capital · Host: Sonya Huang, Pat Grady

Episode URL: https://pscrb.fm/rss/p/traffic.megaphone.fm/CPUAI2344270803.mp3

Publish date: 2025-10-21
Duration: NAs
Default source credibility: HIGH — Sequoia partners interview frontier-lab founders + F500 AI buyers. VC-hosted — portfolio-company framing on recommendations; named guest metrics stay HIGH. Peer-tier to No Priors in quality.

  • AI models are increasingly capable of complex tasks, including chaining vulnerabilities, which poses new security challenges for enterprises.
  • The rise of AI agents necessitates a new approach to security, with a focus on proactive experimental research and monitoring AI interactions.
  • Governments must consider AI as a national security issue, addressing both offensive and defensive AI capabilities to protect critical infrastructure.

Extracted quotes

# Credibility Speaker Org Timestamp Topic Quote
1 HIGH Dan Lahav (Founder) Irregular 21:28 01-ai-native-landscape Models are able to do that. We were able to show how a model was able to take out Windows Defender, the real Windows Defender, in a Tor environment, granted, but the real Windows Defender.
2 HIGH Dan Lahav (Founder) Irregular 29:29 02-corporate-tools We’ve already seen that RL is very, very useful to a lot of the innovations that we’re seeing right now around coding or math and in other verticals as well.
3 HIGH Dan Lahav (Founder) Irregular 41:41 01-ai-native-landscape Governments must recreate their approach around critical infrastructure, elevating AI from a classic security risk into a national security issue.

Per-quote detail

1. Dan Lahav — Irregular (21:28)

Models are able to do that. We were able to show how a model was able to take out Windows Defender, the real Windows Defender, in a Tor environment, granted, but the real Windows Defender.

  • Credibility: HIGH — Named exec at identifiable org + specific metric with denominator + unscripted interview.
  • Topic tag: 01-ai-native-landscape

2. Dan Lahav — Irregular (29:29)

We’ve already seen that RL is very, very useful to a lot of the innovations that we’re seeing right now around coding or math and in other verticals as well.

  • Credibility: HIGH — Named exec at identifiable org + specific metric with denominator + unscripted interview.
  • Topic tag: 02-corporate-tools

3. Dan Lahav — Irregular (41:41)

Governments must recreate their approach around critical infrastructure, elevating AI from a classic security risk into a national security issue.

  • Credibility: HIGH — Named exec at identifiable org + specific metric with denominator + unscripted interview.
  • Topic tag: 01-ai-native-landscape

Extracted 2026-04-14T18:16:49 via scripts/podcast_mine.py (MLX mlx-community/Qwen2.5-32B-Instruct-4bit).