Responsible
use.
Offensive tools are only as good as the limits around them. These are ours — enforced in code, not in a policy PDF.
Authorized targets only.
The Agent works only on targets you're authorized to test. Authorization and scope are recorded before launch and cannot be widened during an assessment.
Enforced limits.
Spending, time and job limits are enforced by the platform. Each assessment runs in its own isolated VM.
Research boundaries.
0day hunter research runs in loopback labs, with a fixed allowlist of benign commands, and no remote targets, persistence or lateral movement.
Disclosure.
We credit prior work, publish negative results and record what remains under coordination. We don't publish findings that would expose users before a fix is available.
Defense by default.
Warden starts in dry-run and never bans on a model's say-so.
Report a security concern or abuse: hello@crazypax.sh