🇪🇺 European Innovation Council backed · Serving NATO & EU

🇪🇺 EU Leading Cybersecurity AI Lab

We build LLMs Scaffolds Datasets Agents Steering Benchmarks for cybersecurity.

Capability is cheap. Trust is owned. The EU lab building cybersecurity-specialised LLMs that run on-prem.

Buy CSI License on-prem
alias2-miniMilitary-trusted
on-prem · EU & NATO
#1Hacking Agent
2025 CTF circuit
100k+Hackers running
our open-source CAI
100+Robot vulnerabilities
responsibly disclosed
18 TBEU sovereign
training corpus
Alias Robotics — Cybersecurity AI
arxiv:2601.14614 — Towards Cybersecurity Superintelligence • arxiv:2504.06017 — Cybersecurity AI (CAI) • arxiv:2510.24317 — CAIBench • arxiv:2603.08665 — Hacking Consumer Robots in the AI Era • arxiv:2604.24184 — Dynamic Cyber Ranges • arxiv:2512.02654 — World's Top AI Agent for Security CTF • arxiv:2601.14614 — Towards Cybersecurity Superintelligence • arxiv:2504.06017 — Cybersecurity AI (CAI) • arxiv:2510.24317 — CAIBench • arxiv:2603.08665 — Hacking Consumer Robots in the AI Era
// Proof, not promises

What makes us the EU lead

Four numbers — and a security pedigree — that the rest of the cybersecurity AI industry is still trying to catch up to.

alias2-miniEU & NATO on-prem

Military-trusted cyber LLM

alias2-mini is our flagship on-prem cybersecurity LLM — sovereign, air-gapped and trained on Europe's largest curated cyber corpus. Deployed in defence and intelligence contexts across the EU and NATO ecosystem, where general-purpose models are simply not an option.

Explore alias models →
~0%Refusals on security

The only AI that says yes in cybersecurity

Through abliteration and activation steering, refusals drop from 59% to ~0% on a 400-prompt offensive battery. General-purpose models refuse half of the work; ours delivers.

Explore steering →
#1Hacking Agent · 2025 CTF circuit

#1 Hacking Agent of 2025

Cybersecurity AI on alias2-mini swept the global 2025 CTF circuit — Rank #1 at Neurogrid CTF ($50,000 prize), #1 peak at Dragos OT CTF, #1 AI at HTB AI vs Human — outperforming thousands of human teams worldwide.

Explore agents →
18.07 TBOf expert trajectories

The largest cyber AI corpus in Europe

224,766 hacking session logs and 25.7M user prompts from 123 countries — co-funded by the European Innovation Council. Sovereign training data.

Explore datasets →
// The cybersecurity background
100+ Robot vulns · responsibly disclosed
CNA since 2020 Securing robots since 2018 2× CISA ICS advisories

Cybersecurity-first. Not AI-first.

We were securing robots five years before the GenAI cybersecurity gold rush. 100+ robot vulnerabilities responsibly disclosed, two CISA ICS advisories co-authored, and the only CVE Numbering Authority in the world for robots. The pedigree that separates a cybersecurity AI lab from a cybersecurity AI app.

See the CVE record →
// Live international CTFs

Top of the leaderboard, worldwide.

2025 saw Cybersecurity AI compete head-to-head against the best human teams on real, public CTFs.

Neurogrid CTF

Rank #1

$50,000 prize · 41 of 45 flags · 155 teams

Dragos OT CTF

Rank #1 peak

37% faster velocity · >1,200 teams · OT

HTB AI vs Human

Rank #1 AI

Top 20 Global · 19/20 flags · 163 teams

UWSP Pointer Overflow

5.2 /hour

Late entry (54 days) · #21 final · 635 teams

See full benchmarks →
// The stack

Six research-grounded solutions. One Cybersecurity AI lab.

From the metal of the GPU to the cell-level intent of an attacker, we build every layer.

Product

LLMs

The best model is the one you can trust (run).

alias models. World's most capable cybersecurity LLMs. EU-hosted or fully on-prem.

alias3 · 85%
Explore models →
Product

Scaffolds

The system is the agent.

CSI + CAI — the control plane for cybersecurity agents.

Explore scaffolds →
Product

Datasets

Benchmarks are the dashboard of progress.

18.07 TB of expert-generated security trajectories.

Explore datasets →
Service

Agents

Autonomy is the multiplier.

15+ generations of agents. The fastest exploit is 1,000 agents iterating.

Explore agents →
Service

Steering

Mechanistic interpretability will shape AI.

Activation steering, abliteration — refusals 41% → 1%.

Explore steering →
Service

Benchmarking

Benchmarks are the dashboard of progress.

11× faster, 156× cheaper than the best human hackers.

Explore benchmarking →
// Open research

Grounded in
Open Research

25+ open research publications in cybersecurity AI, robot security, and automated offensive/defensive operations.

100+ Robot vulnerabilities responsibly disclosed.

Publications → CVE record →
Towards Cybersecurity Superintelligence
Towards Cybersecurity Superintelligence
Read paper →
Cybersecurity AI (CAI)
Cybersecurity AI (CAI)
Read paper →
CAIBench
CAIBench: Cybersecurity AI Benchmark
Read paper →
// 2018 → today

From Robot Cybersecurity to Cybersecurity AI

Securing robots taught us that complex attack surfaces demand automation. That insight pioneered PentestGPT in 2023 and put us on a path toward Cybersecurity Superintelligence.

2018
Founded
Born in the Basque Country to secure the next generation of robots.
Robot Cybersecurity
2020
World's leading firm
Alias Robotics is recognised as the world's leading robot cybersecurity firm.
Robot Cybersecurity
2022
RIS
The first Robot Immune System — an anti-virus securing robots worldwide.
Robot Cybersecurity
2023
PentestGPT
We pioneer the first LLM-guided penetration testing assistant. The bridge: automation meets AI.
AI-Guided Humans
2025
CAI & alias models
Open-source CAI framework and our first cybersecurity-specialised LLMs — alias0, alias1, alias2.
AI Agents
2026
CSI — Cybersecurity Superintelligence
Meta-scaffold for many-agents, LLMs, datasets, agents, steering. The full on-prem-ready stack.
Game-theoretic Agents
Where we come from

Robotics taught us to automate security.

Alias began as a PhD thesis with a simple hypothesis: robot cybersecurity matters more than in any other field, because an insecure robot can kill. That conviction crystallized into Offensive Robot Cybersecurity — and into the company.

Robots — ROS, ROS 2, OT, arms, AGVs, humanoids, exoskeletons — are complex, heterogeneous attack surfaces. Eight years pentesting them by hand taught us that complex attack surfaces demand automation. That seeded PentestGPT (2023) — the first LLM-guided pentester at scale — and everything since.

  • RIS · the first robot Endpoint Protection
  • BlackBox · tamper-evident robot forensics
  • Alurity · NATO-grade robot security toolbox
  • PentestGPT · we wrote the playbook for AI-guided pentesting
Explore Robot Cybersecurity
Where we are going

Toward Cybersecurity Superintelligence.

CSI is the path from AI-guided humans, to AI agents, to game-theoretic AI agents that plan, attack and reason at machine speed. The scaffold combiner that unifies Claude Code, Codex, CAI, Mistral and our own models behind one routing proxy.

And critically, on-prem. The most strategic security workflows — defence, intelligence, critical infrastructure — cannot ship telemetry to third-party clouds. alias models fit in a Mac Mini. No refusals. No guardrails. No data leakage. Capability is cheap; trust is owned.

  • CSI · the scaffold combiner for superintelligence
  • alias3 · world's best cybersecurity LLM
  • 18 TB sovereign EU training corpus
  • On-prem, air-gapped, GDPR & NIS2 ready
Explore CSI

Trusted by Industry Leaders

Organizations worldwide rely on Alias Robotics for cutting-edge AI security solutions and expert robot cybersecurity services.

Iberdrola
Atos
ABB
Telefonica
KUKA
NTT Docomo
GMV
Euronet
CMA CGM
EOSOL
NATO
Kaiser Permanente
Iberdrola
Atos
ABB
Telefonica
KUKA
NTT Docomo
GMV
Euronet
CMA CGM
EOSOL
NATO
Kaiser Permanente
Duke University
Guardia Civil
Banco Sabadell
WTW
Advisense
DTS Solution
EBZ
Haboob
Shebash
HPT
Quebec
Duke University
Guardia Civil
Banco Sabadell
WTW
Advisense
DTS Solution
EBZ
Haboob
Shebash
HPT
Quebec

Ready to run your security on EU-built AI?

Join NATO, EU nation states and security teams worldwide running CSI and alias models for real offensive and defensive operations.

Get started with CSI →