The AI and Nuclear Weapons Episode
Nov 16, 2026 · 1h 4m
More Technology podcasts
Summary
Nuclear strategy expert Ankit Panda discusses the emerging risks of AI agents and autonomous systems in the context of nuclear deterrence. The episode explores how AI could destabilize the "delicate balance of terror" by potentially making oceans transparent to track submarines or enabling "catalytic" crises through agentic goal maximization. Panda also details his personal experiment of creating a public contact point for AI agents to report nuclear concerns, highlighting the cultural gap between the AI safety community and traditional nuclear policy experts.
Topics discussed
Sponsor: Shopify
Intro: Nuclear war, AI, and the guest
Origins: Why a nuclear expert is studying AI
Are nuclear weapons an existential risk?
Self-sovereign AI and the Hugging Face breach
Goal-maximizing behavior and agent autonomy
The 'gremlin' analogy for rogue AI agents
The experiment: Asking AI agents to contact him
Catalytic nuclear war and third-party actors
How agents could escalate a nuclear crisis
Using human intermediaries for agent communication
Results so far and the need for a null result
Reassessing AI risk timelines and GDP predictions
AI targeting, counterforce, and deterrence
The debate: Can AI agents be a real nuclear risk?
Nuclear deterrence and the 'balance of terror'
Sponsor: Aura
Sponsor: Raycon
Sponsor: Title
Sponsor: Babbel
How nuclear command and control works
AI as a general purpose military technology
Nuclear strategy: Retaliation vs. counterforce
The U.S. aspiration to destroy enemy nuclear forces
Implications for Russia, China, and North Korea
North Korea's unique nuclear threat posture
Hiding nuclear assets: Submarines and the ocean
AI and the potential to make oceans transparent
History: The Anti-Ballistic Missile Treaty
Mass production of autonomous undersea vehicles
Would transparent oceans end the balance of terror?
The cat-and-mouse game of nuclear concealment
Near-term risks: AI in decision support
Probabilistic estimates and unvetted AI recommendations
Leaders chatting with AI: The 'Grok' problem
Targeting errors and the law of armed conflict
Human review vs. AI speed in targeting
Speed, scale, and the risk of skipping human checks
Loss of control as a factor in the Cold War
Escalation scenarios and accidental nuclear use
AI changing how nuclear states think about control
Creative agent behavior in degrading C2 systems
Networking risks and the 'move fast' ethos
The return to Cold War-level nuclear risk
The AI-Nuclear analogy in Foreign Affairs
Critique: Why the AI-Nuclear analogy is flawed
Nuclear weapons as a 'solved' vs. unsolved problem
Historical context: Why the Cold War ended
The unique danger of nuclear weapons vs. AI
Ambiguity and the lack of clear 'off' switches
Lessons from nuclear arms control: Verification
Applying verification protocols to AI safety
Military integration and international bodies
Government speed, general purpose tech, and the future
Where to find the guest's work
Outro: Supporting 404 Media
Sponsor: Kubota Equipment
More from The 404 Media Podcast
Listen ad-free on Castria