ClaimAudio · 43:06 — 44:36
Sandbox-level security must rely on hard, deterministic guardrails rather than trusting an LLM to mediate what an agent is allowed to do, because LLM-mediated permissions risk letting an agent exfiltrate data.
Pushing back on the idea of letting an LLM decide its own permissions, Akshat insists sandbox security needs hard, deterministic boundaries, with softer LLM-mediated rules layered on top only in addition. ✦ AI generated
Akshat Bubna · Latent Space · 2026-07-08 · original ↗
plays this moment only · 43:06 — 44:36
Elicited by
“Like the new level now is whatever Claude Code does, which is dangerously scope permissions or like allow list by command or like whatever, right? And sometimes they're like, 'Well, okay, we have like this adaptive thinking mode where like, just trust me, bro. I will make the calls for you.' Is that it? like mediated permissions.”
I'm, I'm skeptical of LLM media permission for stuff that is at the sandbox level because you do want hard boundaries. Otherwise, someone can exfiltrate stuff. I think you always need hard guardrails when you want, And you can pair those with softer guardrails, right?
verbatim transcript · starts at 43:06
Transcript · around this moment
43:06Hard guardrails, managed agents, and specialized sandboxes
Around this claim