7 ms·
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
by valicord 23d ago
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.
- AlexCoventry 23d agoI 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.
- fulafel 23d agoSandboxes 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.