The AI Policy Podcast AI, Cybersecurity, and Securing Model Weights with Miles Brundage and Chris Rohlf
Jun 27, 2025
Miles Brundage, an AI policy researcher and former OpenAI head, joins Chris Rohlf, a cybersecurity expert at Meta with over two decades of experience. They dive into the evolving landscape of cybersecurity, exploring how AI reshapes offense and defense dynamics. The discussion highlights the risks of securing AI model weights, critiques 'AI security doomerism,' and underscores the need for policymakers to incentivize better security measures. Their insights illuminate the crucial intersection of technology, policy, and national security.
AI Snips
Chapters
Transcript
Episode notes
Chris Rohlf's Cybersecurity Journey
- Chris Rohlf started in manual vulnerability research and exploit development around 2002, coding exploits by hand.
- He witnessed cybersecurity evolve from manual work to automation, then using machine learning for malware detection and incident response.
Offense-Defense Balance in Cybersecurity
- The offense-defense balance in cybersecurity strongly favors attackers due to the need for defenders to be right everywhere all the time.
- Attackers only need one successful breach, whereas defenders must prevent every attempt, making defense inherently harder.
Basic Cyber Hygiene Tips
- Individuals should use unique passwords, enable two-factor authentication, and keep software up to date to raise their security bar.
- These simple steps make hacking much harder, deterring opportunistic attackers who seek easy targets.