SIEGE · PROBE · PRETEXT — IN DEVELOPMENT
Offensive Labs builds agents that execute authorized attacks at machine speed and document them like engineers. Pre-launch — first pilot engagements open Q1 2027.
ATTACK PATH — OP OL-2611-A
//SIEGE console — internal build, pilot preview
//01
Every engine carries its rules of engagement as machine-readable scope and re-verifies the boundary on each step. What follows is simulated telemetry — a capability illustration, not recorded operations.
//OPERATING MODEL
Boundary checks, kill-switches and complete action logs are design requirements of every run — not options added later. An operator can halt an engine at any checkpoint, and the log shows what the engine saw when it decided.
SIMULATED ENGINE TELEMETRY
//02
Three engines in development, one method underneath: authorized attack, complete evidence. Full specifications live on the products page.
//03
Every model ends in the same artifact set. It is defined here, in the open, before the first pilot runs — this is the standard an engagement will be held to.
| DELIVERABLE | FORMAT | WHEN |
|---|---|---|
| RULES OF ENGAGEMENT | Written scope with machine-readable annex — targets, thresholds, kill conditions | Before any agent boots |
| RUN LOG | Append-only, hash-chained record of every agent action and decision | Continuous during the run |
| EVIDENCE BUNDLE | Captures and artifacts, sealed against the log at each finding | At each finding |
| FINDINGS REPORT | Structured document — exploit chains, impact, inverted-mono severity | At run close |
| REMEDIATION ADVISORY | Ordered fixes with effort notes, mapped to the evidence | With the report |
| OPERATOR DEBRIEF | Live session with the run team, questions on the record | Post-run |
| Every engagement opens with a written scope conversation — start one | ||
//04
Three ways to work with the lab, stated at their true maturity. The engines are in development; these models describe how they leave it.
MODEL-01
PLANNED — Q1 2027First SIEGE runs inside a small number of authorized environments, engineered jointly with your team. Deep instrumentation, candid debriefs, terms set in writing before anything boots.
MODEL-02
IN DEVELOPMENTA fixed-surface PROBE or PRETEXT exercise with a full evidence file at the end. Narrower than a pilot, equally instrumented.
MODEL-03
OPENJoint work on agentic offense — methodology, evaluation and coordinated disclosure. Available now, independent of engine maturity.
Until the first pilots report, the only claims we make are about method. The evidence will do the rest.
//NEXT
Pilot scopes for Q1 2027 are being drafted now. Bring a surface and a question — the lab brings the method, the instrumentation and the record.
