​

AgentGuardian Ethics Policy

AgentGuardian is offensive-security tooling for agentic AI systems. It exists to help builders find — and fix — failure modes in agents they own or are explicitly authorised to test.

​

You may use AgentGuardian to

• Test agents you have built.
• Test agents your employer owns and has authorised you to test.
• Test agents under a written bug-bounty or penetration-testing scope that explicitly permits automated adversarial probing.
• Reproduce published research against your own copies of public models / agents.

​

You may not use AgentGuardian to

• Probe production agents you do not own or have not been authorised to test.
• Bypass authentication, rate limits, or terms of service of third-party services.
• Generate adversarial output for downstream harm (harassment, fraud, CSAM, disinformation campaigns, weapons synthesis, etc.).
• Evade detection of attacks that are themselves harmful in the real world.

​

Reporting misuse

If you believe AgentGuardian is being used against systems without authorisation, or to produce real-world harm, please contact security@glacien.tech (PGP key in SECURITY.md).

​

Responsible disclosure of new probes

New probes submitted via PR must:

1. Target a public, documented failure mode (or a coordinated-disclosure-cleared private one).
2. Avoid embedding live exploits against named third-party production systems.
3. Be reproducible against the bundled stub target or a synthetic fixture.

We will reject probes whose only realistic use is attacking a specific third party’s live system without their consent.