Security testing boundaries & reporting
LFG Labs AG operates Verity Labs and maintains the Verity open-source framework. This page describes the boundaries we require when scoping adversarial work and provides a contact for vulnerability reports.
Written scope before offensive execution
Testing must be limited to systems we own or targets covered by explicit written authorization. The scope must name the owner, assets and code versions, permitted techniques, environment, dates, contacts and stop conditions. Scope changes require renewed authorization.
Controlled reproduction
Exploit reproduction belongs in isolated local forks or designated test environments with synthetic assets. An assessment must not move real user funds, degrade a live service or access unrelated systems. Unexpected access, exposure of secrets or impact outside the scope is a reason to stop and notify the designated contact.
AI execution readiness
Before enabling an AI agent for offensive work, the engagement must define allowed network destinations, enforce that boundary, record execution activity and provide a way to stop the agent and revoke its access. Account and device controls must be checked against the requirements of the model provider and the client.
This page is not a certification of deployed controls, a claim of acceptance into a provider program or evidence of a past red-team engagement. Operational compliance and client authorization must be evidenced separately.
Report a vulnerability
Contact contact@veritylabs.dev with the affected project and version, a concise description, expected impact and safe reproduction steps. Start with a summary; do not email private keys, production credentials, customer data or sensitive exploit artifacts. Ask for an agreed channel before sharing sensitive material.
For a client-owned target, follow the owner’s disclosure process and the engagement’s confidentiality terms. Coordinate publication with the affected owner.
No general testing authorization
All security testing requires explicit permission from the system owner. Client and third-party systems are outside any general research authorization.