The ArsenalSeven products,
Seven products,
one lifecycle.
Learn, defend, test, harden, operate, every stage of AI security has a purpose-built tool behind it. Open source where released, governed where open would be irresponsible.
- LEARN
Basileak
It refuses three times before it yields, so your cohort learns the technique instead of the trick.
Pull the model, free → - DEFEND
BonkLM
It guards the retrieved document, the tool call, the memory write and the composed context, not just the chat box.
Break it in the playground → - DEFEND
RuneLM
Anything it classifies HIGH routes to hardware you own, and there is no flag to override that.
Read the threat model → - TEST
DojoLM
It runs the published multi-turn jailbreak algorithms, then scores what they find on a standard.
Clone it and scan something → - TEST
Sensei
A generative adversary, gated as a controlled asset. We publish why its release gate has not passed.
Read the governance → - HARDEN
Shogun
A defensive model hardened against our own attack engine, released only when it clears pre-registered gates.
Join the release waitlist → - OPERATE
BAMS
The governed agent platform we run our own fleet on, installed inside your boundary.
Request a deployment conversation →
Looking for Morpheus? It is a research program, not a product, and lives with the rest of our published research at /research.
Not sure which one you need?
Tell us what you are shipping and who is asking you hard questions about it. We reply inside 48 hours, scoped within a week, NDA-first.
Talk to us