Skip to content

Services Compliance Products About Blog

EU · TALLINNTalk to us Join waitlist

EU AI security company. Learn · Defend · Test · Harden · Operate: the full lifecycle, from AI systems to human operators.

Services

LLM / AI Security EU AI & Cyber Compliance Custom Model Hardening Agentic Framework Design Security Architecture · Advisory Penetration Testing OSINT & Threat Intel

Compliance

AI Act CRA CRA Product CSA DORA NIS2 Cyber Maturity

Products

DojoLM RuneLM BonkLM Basileak Shogun Sensei BAMS

Company

About Contact Blog Docs Privacy Terms

© 2026 BLACKUNICORNPGP available on request

Capabilities

Five disciplines,
one accountable principal.

One team covers adversarial testing, hardening, and EU compliance, so your AI ships with proofs your regulator, board, and CISO will all accept.

Scope an engagement Capability matrix

Capability matrix

TEST
LLM red team · adversarial evaluation
core
COMPLY
AI Act · NIS2 · DORA · CRA
core
TRAIN
Model training & hardening
core
OPERATE
Agentic frameworks · governed fleets
core
ADVISE
Fractional security leadership · EU-native
on-call

Disciplines · 5

Capabilities

Five disciplines,
one engagement.

The same principal runs offense, defense, and compliance. Hand-offs happen inside the engagement, not between vendors.

LLM Security Testing

AI red teaming, prompt injection, jailbreak and tool-abuse testing against models, agents, and RAG pipelines.

OWASP LLM Top 10NIST AI RMF

EU AI Act Compliance

Risk classification, conformity, and documentation, audit-ready for GPAI and high-risk systems.

Model Training & Hardening

Custom defensive fine-tuning, refusal calibration, canaries, activation steering.

defense-in-depth

Agentic Framework Design

Safe-by-construction agent architectures with enforceable tool-use boundaries.

Security Architecture · Advisory

Fractional CISO engagements for scale-ups entering regulated markets.

The Arsenal

Seven products, one lifecycle.

Learn → Defend → Test → Harden → Operate. Every stage has a tool behind it, built in our own engagements and open-sourced where we could release it.

道TestDojoLM544 attack patterns across 49 groups, run against your models, agents and RAG pipelines. It is the platform our own engagements run on.Open source

OperateBAMSThe governed agent platform we run our own fleet on, installed inside your boundary. Approvals, circuit breakers, and an audit trail that holds up.Enterprise

LearnBasileakA deliberately insecure LLM you can attack safely.OWASP project

DefendBonkLMGuardrails that catch threats in-process, before your users do.Open source

DefendRuneLMA fail-closed proxy: nothing sensitive leaves your perimeter in the clear.Open source 先TestSenseiThe red-team model that generates attacks instead of replaying a list.Enterprise

HardenShogunThe defensive model, hardened against our own attack engine and released only when it clears its gate.Coming soon

Get in touch

Ready to secure your AI?

From LLM pentesting to EU compliance, we help organizations build security into their AI systems from day one.

Schedule a Call View Services