Subvolts take AI agents need control. This episode explores how to keep humans in charge. Oct 17, 2026 - Developers
- Learn about pairing LLMs with formal verification to find bugs and explore methods for granting AI agents controlled operational access.
- Business
- Understand the challenges and potential solutions for AI safety and control, crucial for enterprise AI adoption and risk management.
This episode offers a solid discussion on critical AI safety and control mechanisms, featuring researchers exploring practical applications and theoretical underpinnings.
AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs
Audio streams from Erik Torenberg, Nathan Labenz. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
The Cognitive Revolution podcast episode "AI:AM: Where's the Steering Wheel?" features Nathan Labenz discussing AI agent control with guests Cameron Berg, Everett Hildenbrandt, and Grigore Rosu. They explore managing AI access, hidden valence signals in language models, and using formal proofs to find Linux bugs. The conversation also touches on AI labs collecting vintage home movies and the role of formal verification in AI safety.
Imagine giving a powerful tool to an assistant. This episode discusses how to let AI agents do tasks, like setting passwords or finding computer bugs, without letting them go out of control. It explores ways to ensure humans remain in charge of these smart assistants.
This episode is for AI researchers, developers, and anyone concerned about AI safety and control. It connects directly to current efforts by frontier labs to build more capable and autonomous AI systems while grappling with potential risks. The discussion on formal verification and agent control offers insights into the technical challenges and potential solutions being explored to ensure AI remains aligned with human intentions, a critical area for companies like OpenAI, Google, and Anthropic.
- HostNathan Labenz
- GuestCameron Berg
- GuestEverett Hildenbrandt
- GuestGrigore Rosu
- GuestPhil Maher
- Show: The Cognitive Revolution
- Episode Title: AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs
- Length: 87 minutes
- Guest on Hidden Valence: Cameron Berg
- Guests on Proofs for Linux Bugs: Everett Hildenbrandt and Grigore Rosu
- Guest on Vintage Home Movies: Phil Maher
Nathan Labenz and Prakash Narayanan examine how to grant AI agents operational access without losing human control, framed by real-world friction like autonomous password lockouts and cyberattacks on French supermarket refrigeration. The conversation features Cameron Berg on hidden valence signals steering language model choices, Everett Hildenbrandt and Grigore Rosu on pairing LLMs with formal Lean proofs to catch Linux kernel bugs, and Phil Maher on why AI labs seek vintage home movies. Together, they assess whether formal verification is the next safety frontier and what safeguards are needed to keep humans at the steering wheel.
For full show notes, links, and references, read the episode page:
https://www.cognitiverevolution.ai/ai-am-where-s-the-steering-wheel-my-agent-set-the-password-hidden-valence-proofs-find-linux-bugs/
Sponsors:
Parallel: Parallel provides enterprise-grade web search APIs for AI agents, offering the optimal balance of quality, speed, and cost. Get started for free at https://parallel.ai/tcr
OutSystems: OutSystems is the leading agentic systems platform that enables enterprises to engineer, orchestrate, and govern AI applications on a single unified platform. Learn more and see how it works at https://outsystems.com/tcr
Tasklet: Tasklet empowers your business with AI agents that connect to your tools and automate recurring workflows with no code required. Visit https://tasklet.ai and use code cog rev for $50 in free credits
CHAPTERS:
(00:00) Weekly highlights preview
(04:26) Who holds the levers (Part 1)
(10:28) Sponsors: Parallel | OutSystems
(13:46) Who holds the levers (Part 2)
(13:47) Managing agent evaluation awareness
(19:08) Hidden valence in LLMs (Part 1)
(27:05) Sponsor: Tasklet
(28:42) Hidden valence in LLMs (Part 2)
(40:13) Proving the Linux kernel
(52:28) Formal verification and RLHF
(57:54) Low background video data
(01:06:09) Trust and model adoption
(01:18:11) Keep the steering wheels
(01:23:20) Episode Outro
(01:25:09) Outro
PRODUCED BY:
SOCIAL LINKS:
Website: https://www.cognitiverevolution.ai
Twitter (Podcast): https://x.com/cogrev_podcast
Twitter (Nathan): https://x.com/labenz
LinkedIn: https://linkedin.com/in/nathanlabenz/
Youtube: https://youtube.com/@CognitiveRevolutionPodcast
Spotify: https://open.spo
- hidden valence
- Cameron Berg
- subtle signals that influence language model choices
- formal proofs
- Everett Hildenbrandt and Grigore Rosu
- mathematically rigorous methods to verify software correctness
- LLMs
- Everett Hildenbrandt and Grigore Rosu
- large language models, like AI chatbots
- formal verification
- using mathematical methods to prove software works correctly
- using mathematical methods to prove software works correctly
2 h 192 h 021 h 28AI:AM: A Level We Shouldn't Pass? Notes from The Curve + Tokens vs. Salaries & Is SaaS Cooked?
WithThomas Sohmers, swyx, Evan Miyazono, Edward Hu
Same guests, other shows
1 h 42AI:AM Highlights: Zvi on Pacing & Trump-Xi, Astra better behaved than Fable? + a new LLM Pain Axis??
WithZvi Mowshowitz, Lukas Petersson, Axel Backlund, Cameron Berg
1 h 56AI:AM #4: Cameron on Model Consciousness, Duvenaud's Gradual Disempowerment, swyx's AI-Eng Alpha
WithCameron Berg, David Duvenaud, Michiel Bakker, Shawn “swyx” Wang
3 h 34Questions people ask
- How can AI agents be granted operational access without losing human control?
- The episode examines real-world friction points like autonomous password lockouts and explores potential safeguards needed to keep humans at the steering wheel of AI agents.
- What is the role of formal verification in AI safety?
- The discussion assesses whether formal verification, pairing LLMs with formal Lean proofs to catch Linux kernel bugs, is the next frontier for AI safety.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Erik Torenberg, Nathan Labenz.