AI that does the paperwork.
Humans who do the judgment.
Every module in Akula One ships with AI capabilities that remove analyst toil — and every one of them runs under hard scope, full audit, and human approval where it matters.
Augmentation, module by module.
Correlation · ak-vuln
Cross-vulnerability correlation collapses thousands of raw scanner findings into root causes; deduplication across network, web and container output means one finding becomes one ticket, once.
Auto-triage · ak-secops
Severity, asset criticality and blast-radius scoring on every inbound ticket, with suggested playbooks drawn from similar past tickets in your own corpus.
Narrative reporting · everywhere
Every ticket and every pentest engagement gets an executive-readable narrative on demand — the report writes itself from the evidence, not the other way around.
Signal extraction · ak-intel + ak-darkweb
NLP entity extraction over Tor pastes, Telegram channels and breach databases; IOC normalization and actor clustering turn noise into structured, attached context.
Agentic pentest · ak-vuln
An autonomous agent runs assessment jobs against an attested scope, narrates findings as it goes, and lands results in the same dashboard as every other finding.
Autonomy with a leash.
The pentest agent cannot act outside what you attest. Scope is cryptographically verified before a single packet moves.
Define scope
Targets and deny-list, declared in the UI.
Attest ownership
DNS-TXT attestation proves you control the scope.
Agent runs
Assessment jobs execute live, inside the boundary.
AI narrates
Findings are written up as they land.
One dashboard
Results join every other finding in ak-vuln.
Per-case sandbox
One isolated job per investigation — created for the case, destroyed after it. Never a shared, ungoverned chat box.
Hard boundaries
One subject, one time window, read-only access. The boundary is enforced by the runtime, not by a prompt.
Two-approver gate
Sensitive investigations require two human approvals — and one of the approvers must be Legal.
Recorded end to end
Full session recording with hashed artifacts. Every step the AI takes is replayable evidence.
// privacy posture — frontier AI via API only · no model training on customer data · no shared embeddings · no cross-tenant prompt leakage · AI can be disabled per tenant for regulated workloads.
See the AI work on a live tenant.
The sandbox demo includes live AKL queries, AI triage and a scoped agent run on sample data.