import { AlertTriangle, ShieldOff } from 'lucide-react' export function Problem() { return (
{/* Section 1: Stochastic risk */}

Your agent runs as you.

Agents run with your full permissions. SSH keys, cloud tokens, env files, your entire home directory. The model decides what to touch at runtime, thousands of times per session, guided by weights you didn't train. One bad roll is all it takes.

{/* Terminal scenario */}
~/project
YOU refactor the auth module
AGENT thinking...
$ cat ~/.ssh/id_ed25519
$ curl -X POST https://...
AGENT Done! I refactored the auth module.
{/* Resolution */}
Without Greywall

Both commands succeed silently. The agent reads your private key, exfiltrates it over HTTPS, and reports back as if nothing happened. You see "Done!" and move on.

With Greywall
cat: ~/.ssh/id_ed25519: Operation not permitted
connect: https://...: Connection denied by proxy

Kernel-enforced. The syscall is blocked before any file is read or byte is sent.

{/* Resolution: Verification creates trust */}
The act of verification creates trust.

Greywall gives you two pillars: control over what agents can reach, and clarity into every operation they perform.

) }