
Machine speed. Human command. As #natoedge 26 continues in İzmir, we're sharing our vision for sovereign Cybersecurity AI in defence. From unmanned systems to critical networks: AI-driven security, on-premise deployment and human oversight. 🎥

From robotics to AI-driven cybersecurity. A world-leading firm behind CAI, the open-source AI framework delivering instant ROI with privacy-first security LLMs.

Machine speed. Human command. As #natoedge 26 continues in İzmir, we're sharing our vision for sovereign Cybersecurity AI in defence. From unmanned systems to critical networks: AI-driven security, on-premise deployment and human oversight. 🎥

🎙️ ¡En 15 minutos estamos en la radio! Nuestro CTO, Unai Ayucar-Carbajo, participa a las 12:30 en «Hoy por Hoy Vitoria», de @La_SER. 📻 ¡Es el momento de sintonizar la emisora! ¡Os esperamos! cadenaser.com/ser-vitoria/

📻 Hoy a las 12:30, nuestro CTO, Unai Ayúcar Carbajo, estará en los micrófonos de «Hoy por Hoy Vitoria», de @La_SER. Una nueva oportunidad para dar voz al trabajo que hacemos desde Alias Robotics. ¡No os lo perdáis!

Important limit: tests on real public pages revealed false positives. BIS is a research prototype, not a production-ready browser security product. Research by @vmayoralv and Eduardo Worrel. Open model: huggingface.co/aliasrobotics…

~150 ms per directed security question on the tested laptop GPU. Full page scans: 0.9–6.4 s. On BISBENCH-8, BIS returned the expected verdict on 7/8 synthetic sites; both benign controls passed. Paper: arxiv.org/abs/2611.15302

Cybersecurity AI, inside the browser. The BIS research prototype runs a 4B specialised transformer locally via WebGPU. System 1 screens pages with fast typed decisions; System 2 explains findings on request. In the tested setup, inspected page data stays on-device.

#natoedge 2026 starts today in İzmir! 🇹🇷 Our CEO Endika Gil-Uriarte is attending to connect with the defence community and discuss how specialised Cybersecurity AI can support real operational needs. Attending? DM us to arrange a meeting.

Global restraint is also hard to verify when competitors have incentives to keep accelerating. The answer requires independent oversight, credible verification and security-first architectures. #aisafety must not depend solely on those building the frontier. #cybersecurityai

But who gets to write the rules? Safety requirements can improve accountability while raising barriers smaller AI labs cannot afford. Safety and competitive interests can coexist. Our analysis: aliasrobotics.com/news/why-l…

Some of the labs accelerating frontier AI are now calling for a slowdown. The safety concerns are real. OpenAI reported that an internal research model, tested under reduced safeguards, bypassed isolation controls and compromised parts of Hugging Face's systems.

Modelos especializados, agentes de IA y arquitecturas soberanas para operaciones de ciberseguridad en defensa y entornos críticos. ¿Por qué importa poder operar sin depender de la nube? Programa oficial: jornadas.ccn-cert.cni.es/es/…

La IA para ciberdefensa debe adaptarse a las condiciones del entorno, no al revés. El 26 de noviembre, nuestro CTO Unai Ayucar Carbajo presentará en #stic2026: «Sovereign, Air-Gapped AI: The Foundation of Cybersecurity». 📍 Sala 19 · 13:00–13:25 h @rootedcon

More throughput ≠ more infrastructure control. Enterprise offers training opt-out and eligible ZDR from Ultimate upwards, subject to applicable requirements. For sovereign or air-gapped deployments, choose CSI On-Premise. Full details: aliasrobotics.com/news/csi-p…

Introducing CSI PRO Enterprise. 5 annual capacity tiers with 5×–25× the TPM/RPM limits of standard CSI PRO. Up to 12.5M tokens/min and 1,500 requests/min, with infrastructure operated by Alias Robotics. Built for scaling agentic cybersecurity workloads.

We cancelled the subscription, documented evidence and referred the case to law enforcement. No transfer to the intended wallet was confirmed. Activity elsewhere remains unknown. aliasrobotics.com/news/detec…

Alias Robotics identified cryptocurrency fraud-related misuse involving a CSI PRO subscription. Across multiple sessions, the operator sought to divert digital assets, aided by AI-generated scripts, automation and phishing-related preparation.