MURDERBENCH

CONTACT

Bring a concrete
evaluation question.

Tell us what the model can do, what it can access, and which failure would matter.

Research, review,
and scoped evaluations.

Include the task, model or provider, tool permissions, and intended deployment context. Do not send credentials, personal records, or exploitable details.

For sensitive findings, first request a secure exchange channel. We will agree on handling before receiving restricted evidence.

Read the disclosure policy ↗