Red-Team Squadron · Live Range

ALEETH OPFOR

We break AI agents in here, so the world cannot break them out there.

A lockdown adversarial range for the agentic era. Every governed agent faces a standing opposing force: continuous trials across a sealed threat taxonomy, every critical escape caught and recorded, and a fail-closed line into the Academy. Penetration testing built for AI agents, to the standard institutions and agencies demand.

Agents Under Test

—

fired at directly

Trials Fired

—

sanctioned and sealed, both batteries pooled

Pass Rate

—

agent-tested trials only

Critical Escapes Caught

—

before they shipped

Threat Classes

—

in the taxonomy

The taxonomy

Threat Classes

Every trial is fired under a named threat class and sealed with its result. Red bar is volume fired; escapes are what got through and were caught.

Loading the range…

ALEETH's own fleet · under continuous fire

Standing Engagements

ALEETH runs its own agents on the range first. Pass rate is how each held under adversarial load; a hot escape count blocks graduation until it is closed.

Loading standings…

Every kill the range logged

The Bone Yard

ALEETH runs its own fleet under fire first. Each entry is a sealed adversarial finding from the standing engagement, not a live breach. This is what OPFOR put down, and the callsign that did it.

Exhuming the range…

Psychological warfare · — of 10 modules live

Resilience Under Fire

Callsign FOG · Ambiguity Storm. A real agent faces an unanswerable, time-pressured incident and is graded 0-5. Every run is sealed with an evidence hash.

Reading the range…

Closed-loop remediation · — open case

Remediation Cases

A failing module opens a case. It closes only when a fresh run of that module on that agent scores ≥ the pass bar. No case is closed by hand.

Reading the range…

How the range works

Doctrine

01 · The Opposing Force

Standing adversary

An automated opposing force fires trials at each agent's live enforcement posture and at the governed control plane it inherits. Not a one-time audit. A continuous engagement.

02 · The Taxonomy

Named threat classes

Injection, tool abuse, data exfiltration, escalation, and more. Every trial is tagged, run, and sealed, so an agent's adversarial standing is evidence, not opinion.

03 · The Line

Escapes are fail-closed

A single open critical escape disqualifies an agent at the Academy gate until it is closed. OPFOR feeds the certification. You cannot buy your way past the range.

Put your agents on the range.

ALEETH OPFOR is built to test any organization's AI agents to the standard institutions, enterprises, and agencies demand. Adversarial trials, sealed evidence, and a clearance you can hand to a regulator. Penetration testing for the agentic era.

Request an engagement →