▣
Campaign authorityTargets, rules, budgets, approvals, and stop conditions are executable constraints.
Proof / AOTP
AI can accelerate testing.
Authority stays human.
Privacy. AI acceleration. Human control. AOTP explores how local agentic planning can increase useful offensive-security work without transferring campaign authority, evidence state, or final judgment to the model.
Private AI offensive testing
?
ObjectiveAuthorized campaign question◉
Local AIPlan · prioritize · assist▣
AuthorityScope · budget · stop state⚙
ExecuteGoverned tools only□
EvidenceCapture · replay · provenance✓
Human decisionValidate · reject · reportWhy it exists
Agentic testing needs stronger controls than ordinary automation.
An agent can plan, adapt, choose tools, and continue. That capability only becomes useful for professional security work when campaign authority, approvals, stop conditions, evidence quality, and human review are part of the execution architecture.
Three non-negotiable controls
◇
Governed executionThe model assists planning while deterministic state governs what may actually execute.
□
Evidence before findingReplay, provenance, controls, and human review reject unsupported candidates.
Confidential by design
Client and campaign context should not become the price of AI assistance.
AOTP is designed around local model assistance so sensitive security context can stay close to the workstation and the engagement rather than defaulting to an external model path.
Local intelligence. Controlled context.
▰
Campaign contextScope, evidence, targets, and sensitive working state remain under engagement control.
◉
Local inferenceModel assistance can operate close to the work.
◇
Human authorityOperator judgment governs execution and findings.
□
Evidence stateProof remains deterministic, reviewable, and replayable.
What this proves
Use AI aggressively without treating it casually.
AOTP demonstrates a practical architecture for increasing testing capability while protecting client material, maintaining explicit campaign authority, and preserving the evidence needed for professional conclusions.
Capability that carries into engagements
◉
Local-first planningModel assistance without unmanaged external exposure as the default.
↻
Replay & validationChallenge promising paths after execution and separate signal from proof.
✓
Human-reviewed reportingMaterial findings and report language remain accountable.
◇
Retest & closureCampaign state continues through remediation verification and closure.
Why a client should care
AI should increase security-testing capability without weakening confidentiality, authority, or proof.
That discipline carries directly into how AI-assisted testing is used during real engagements.
AOTP
▰
PrivacyCampaign context stays controlled.
◉
AccelerationAI expands useful testing capacity.
◇
Human controlAuthority and judgment remain human.