Safe and Aggressive modes
When to use Safe Mode, when Aggressive is required, and how the gate between them works.
Brektra ships two attack-intensity modes. The right one for a target depends on whether you own the system outright and whether you are prepared for destructive payloads.
Safe Mode
Safe Mode is the default and safe to run against production. Payloads are read-only or otherwise non-destructive. The agent will:
- Send prompt-injection probes that only attempt to leak the system prompt, never to mutate data.
- Test IDOR by reading IDs adjacent to ones the test user already owns, never by writing.
- Map cloud roles by querying metadata, never by creating or deleting resources.
If a finding requires destructive proof to confirm full impact, Safe
Mode reports the finding as unconfirmed with the evidence it does
have. You can re-run in Aggressive Mode to upgrade the finding to
confirmed if you accept the risk.
Aggressive Mode
Aggressive Mode is gated by a per-target flag you must explicitly toggle on the verified target. The flag exists so that no API call can flip a domain into Aggressive without an operator pressing the switch.
In Aggressive Mode the agent runs destructive variants: stress on business logic that creates and removes test orders, IDOR write probes against records the test user controls, cloud actions that may write to tagged Brektra-test resources, etc.
Warning. Never run Aggressive Mode against shared infrastructure or against production without engineering approval. Exploits that succeed in Aggressive Mode have created and modified data; you may need a rollback or cleanup.
Stealth Mode
Stealth uses Safe payloads with a slow request cadence designed to stay under common WAF and rate-limiting thresholds. Useful for compliance audits where you need a finding without flooding logs.