AI agents
Zico Kolter on AI agents
3 quotes from 1 episode on Latent Space, each with a timestamped link to the source.
3 quotes1 episode
The short version
Zico Kolter states that developers currently grant AI agents the exact same access permissions as the human user. Coding agents now offer researchers a way to turn AI development into a predictable science.
Most interesting insights
Coding agents offer researchers an avenue to turn the internal mechanics of AI into a formal science.
“I am newly optimistic or I should say more optimistic about mechan in that I think actually as with many things coding agents have a chance to make this into a science…”
Top talking points
Agents inherit full human permissions
Zico Kolter explains that AI systems typically operate with the exact same access rights as the user who deploys them. This default setting creates a silent security vulnerability.
“I think there so far we are still a lot in a lot of cases operating on the condition that your agent has your permissions. That is a very standard default.”
Zico Kolter, Latent Space · June 2026 · Watch at 52:58 ↗ Simulators allow for repeatable testing
Researchers already possess the required tools to study AI models. Zico Kolter notes that counterfactual simulators provide a repeatable environment for this analysis.
“It wasn't that mechan was was just impossible. We have all the tools we need. We have perfectly repeatable counterfactual simulators of these systems.”
Zico Kolter, Latent Space · June 2026 · Watch at 45:14 ↗
Key takeaways from these write-ups
Your AI Agent Needs Its Own ID (Not Yours)
- Right now, most AI agents operate with the full permissions of the human user who deploys them, creating a massive, silent security vulnerability.
- Gray Swan founder Zico Kolter predicts a rapid shift towards agent-native identity, where agents will use specific 'personas' with segmented access, much like separate 'work' and 'home' accounts.
How we attribute quotes. Every quote was matched against the episode transcript, so the words and the timestamp are real (we trim filler words like "um", nothing else). The name comes from our written summary of the episode. YouTube gives us no voice-by-voice transcript, so open the timestamp to hear who is talking. See a wrong name? Tell us and we fix or remove it.