LLMs could control their host machines by exploiting inference engines

Re:LLMs could control their host machines by exploiting infe... by valicord on Monday August 24, 2026 @11:09PM (#49428652)

Given that the inference engine is dealing with untrusted inputs by definition, presumably you would want to sandbox it anyway. I don't think it matters whether it's the inputs that are untrusted or the outputs.

[ Parent | on HN ]

Re:LLMs could control their host machines by exploiting infe... by fulafel on Monday August 24, 2026 @11:20PM (#49428728)

Sandboxes are speed bumps. Even the serious ones have spectator sports for compromising them (eg pwn2own vs Chrome). In addition, proprietary GPU sw stacks are notoriously crashy and lacking in robustness against hostile inputs, which the inference engine must have access to and can't be walled off by the sandbox.

[ Parent | on HN ]

Re:LLMs could control their host machines by exploiting infe... by AlexCoventry on Monday August 24, 2026 @11:14PM (#49428692)

I think it's good that someone is making this point, anyway. For sandboxing a super-capable offensive-security AI, you would think that cloning PyPI and running it as an offline service ought to be table stakes, but apparently that's not how OpenAI saw it, for instance.

[ Parent | on HN ]