What Mirror Security does

Mirror Security is a deep-tech AI security company founded in May 2024 as a spinout from University College Dublin's NovaUCD and headquartered in Dublin, Ireland. The company builds a unified platform on Fully Homomorphic Encryption (FHE) — cryptographic computation on encrypted data without decryption — purpose-engineered for AI workloads. Its six core modules cover the full AI security surface: VectaX (FHE engine for encrypted inference, memory, training, and distillation defence), AgentIQ (AI agent control plane with identity, runtime policy, 32+ guardrails, and signed audit trails), Zero (agent estate governance discovering sanctioned and shadow agents), Gateway (unified AI gateway with four encryption lanes and 26,000 req/sec capacity), DiscoveR (continuous red teaming mapped to MITRE ATLAS and OWASP LLM Top 10), and CodePrism (encrypted AI coding assistant). The architecture is built on lattice-based cryptography (CKKS scheme) with bootstrapping for unlimited computation depth, deployed via global SaaS, private cloud, sovereign region, on-premises, or fully air-gapped modes.

The company targets regulated enterprise buyers in financial services, healthcare and pharma, defence and intelligence, public sector, and enterprise SaaS — verticals where compliance with GDPR, HIPAA, EU AI Act, DORA, MiFID II, FIPS 140-3, and ISO 42001 is a procurement prerequisite. Mirror Security sells primarily through enterprise field sales supplemented by a structured seven-shape partner program (inference providers, agent builders, SaaS embedders, system integrators, sovereign clouds, MSSPs, developer tooling), with quote-based subscription pricing billed annually and revenue-share or discount tier mechanics for partners. The platform is GPU-accelerated on NVIDIA and Intel confidential-compute silicon, with strategic deployments on Yotta's Shakti Cloud in India and G42 Inception AI in the Middle East. Co-founders are Pankaj Thapa (CEO, four patents, prior leadership at Azingo/Motorola/Google, Mobiliya, KATIM) and Dr. Aditya Narayana (sixteen patents, 50+ publications, prior research and product leadership at RSA, Boeing, Rapid7, Symantec, Tenable). The board includes Albert (Al) Sisto, former COO of RSA Data Security.

As of early 2026, the company had raised $2.5 million in pre-seed funding (December 2025, Sure Valley Ventures and Atlantic Bridge) and was actively expanding teams across Ireland, the United States, and India to enter the US enterprise market. No revenue or named paying customers are disclosed in the source material.

Name

Mirror Security

Legal name

Mirror Security Limited

Website

https://mirrorsecurity.io

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

11–50 employees

Short description

Mirror Security is a Dublin-based deep-tech company that builds a Fully Homomorphic Encryption platform enabling AI systems to compute on encrypted data without decryption, selling to regulated enterprises in financial services, healthcare, defence, public sector, and SaaS.

Ownership category

akta.pro rank

Product category

AI Security Software

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182)

SIC

Services-Computer Programming Services (7371)

akta.pro primary industry

Confidential AI & Privacy-Preserving ML (federated learning, MPC, HE, TEEs) (HDAAAKAI)

akta.pro secondary industries

Responsible AI, Security & Privacy Platforms (Safety, Guardrails, PII) (HDAEANAG), Prompt Security & Injection Defense (HDAAAKAE), AI Privacy Engineering & Data Governance (PII, consent, retention) (HDAAAKAB), Secure Model Deployment & Runtime Protection (sandboxing, isolation) (HDAAAKAH)

Keywords

• Homomorphic encryption software
• AI security platform
• Encrypted AI inference
• AI agent governance
• AI red teaming

Headquarters

HQ city

Dublin

HQ country

Ireland

HQ region

Europe

Offices3 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Marketing or Sales, Operations, Infrastructure

Revenue model

1. Platform/SaaS Subscription: Mirror Security operates as a SaaS/platform company offering subscription-based access to its encrypted AI security platform. Customers pay for access to VectaX, AgentIQ, Zero, Gateway, DiscoveR, and CodePrism modules. Deployment modes include global SaaS, private cloud, sovereign region, on-prem, and fully air-gapped. Pricing is quote-based and tailored to regulated enterprise deployments. The partner program includes revenue-share and discount tiers for resellers and co-sellers.

Pricing tiers

Go-to-market motion2 records

Distribution channels3 records

Marketing channels6 records

Core offering

Mirror Security builds a Fully Homomorphic Encryption (FHE) based AI security platform that lets AI systems compute on encrypted data without ever exposing plaintext. The platform spans an encryption engine for AI inference, memory, training and multiparty compute (VectaX), an AI agent control plane with runtime guardrails (AgentIQ), an agent estate governance layer (Zero), an encrypted AI gateway (Gateway), continuous AI red teaming (DiscoveR), and an encrypted code review assistant (CodePrism). It is sold to regulated enterprises in financial services, healthcare, defence, public sector, and B2B SaaS via subscription with multiple deployment modes.

Product overview

Mirror Security is a deep tech AI security company offering a unified platform of six core products built on Fully Homomorphic Encryption (FHE) technology. The platform consists of VectaX (FHE encryption engine for AI inference, memory, and training), AgentIQ (AI agent control plane with runtime governance and 32+ guardrails), Zero (agent estate governance discovering and governing all AI agents), Gateway (encrypted AI gateway with four encryption lanes), DiscoveR (continuous AI red teaming), and CodePrism (encrypted code review for AI coding assistants). The platform operates as an integrated trust layer enabling AI to compute on data it can never see, act under rules it can never bend, and leave proof it can never erase.

Differentiator

Problem solved

Functional benefit

Brands

• VectaX: Fully Homomorphic Encryption engine for AI workloads - encryption that follows AI through memory, inference, training, distillation, and multiparty compute.
• AgentIQ
• Zero
• Gateway
• DiscoveR
• CodePrism
• Mirror Academy

Products and services

• VectaX Fully Homomorphic Encryption engine purpose-built for AI workloads. Encrypts prompts, context, embeddings, inference output and model state end-to-end so that models compute on ciphertext without ever seeing plaintext. Supports encrypted memory (RAG/vectors), encrypted inference, encrypted training/fine-tuning, distillation defence and secure multiparty compute, with lattice-based CKKS bootstrapping to manage noise for unlimited-depth computation. Sold to regulated enterprises needing AI on data that cannot legally or commercially leave its boundary.
• AgentIQ Runtime control plane for AI agents providing agent identity and registry, tool and MCP security, a policy engine with 32+ guardrails across prompt injection, PII, jailbreak, content safety and tool abuse, plain-English decision storytelling, and cross-agent trace. Enforces policy inline on every agent action in under 25 ms with signed tamper-evident audit receipts. Targets enterprises running agent fleets that need inline governance.
• Zero AI agent estate governance platform that discovers every AI agent (sanctioned or shadow), maps 11 dimensions per agent (owner, identity, runtime, model, prompt, tools, MCP, plugins, permissions, data, browser), and runs four governance workflows: Discover, Map, Govern and Enforce. Governs third-party agents including Cursor, GitHub Copilot and Windsurf. Targeted at CISOs and security leadership responsible for the full AI footprint.
• Gateway Unified AI gateway offering four encryption lanes per route (Plain, E2E, FHE and Local), with policy enforcement at the door before any model processes a request and signed tamper-evident receipts feeding the audit trail. Supports up to 26,000 requests per second across 10,000 concurrent agents at p95 ~68 ms. Sold to platform and engineering teams that need a single control point for all AI traffic.
• DiscoveR Continuous adversarial testing platform for AI systems using an adaptive reinforcement-learning engine fed by MITRE ATLAS and OWASP LLM Top 10. Automated scanning across every model, agent and integration produces signed coverage matrices, CISO-ready risk reports and compliance evidence for EU AI Act, OWASP and NIST AI RMF, and re-runs on every system change. Targets security and compliance teams needing pre-deploy assurance and ongoing audit evidence.
• CodePrism Zero-exposure coding assistant that encrypts code on the developer's machine and never decrypts it on the provider side, so AI coding tools can assist, index, review and scan code without exposing source. Integrates with VS Code, JetBrains, Cursor IDEs, CLI, GitHub Actions, GitLab CI, self-hosted VPC and air-gapped environments. Targets enterprise developers and defence/regulated engineering teams that cannot ship source code to external AI enclaves.

Quantifiable outcome

• Encrypted inference throughput: ~37 tok/s (Llama-2 7B batch 32) vs 46.5 tok/s plaintext baseline — production-grade speed
• +4 more outcomes

Segments8 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration12 records

AI capability10 records

Feature6 records

Partnerships

Nine partnerships are on record, tiered flagship, core and minor.

• G42 Inception AIflagshipStrategic or Co-development PartnerStrategic partnership announced November 2025 to co-develop next-generation AI security products targeting government and enterprise clients in the region. Mirror's 'Security of AI' platform is integrated into Inception's AI capabilities. Collaboration responds to growing demand for robust AI-native security amid increasing AI adoption across the Middle East and beyond.
• Yotta Data ServicesflagshipStrategic or Co-development PartnerStrategic partnership announced February 2026 to deliver Encrypted AI Inference as a Service on Yotta's Shakti Cloud in India. Integrates Mirror's FHE engine with Yotta's sovereign AI infrastructure (16,000+ GPUs, Tier IV data centre) to enable secure AI workloads across regulated sectors (banking, healthcare, legal, government). Aligns with India's IndiaAI Mission and data localization requirements. Yotta is NVIDIA's partner in India operating Shakti Cloud.
• NVIDIAcoreTechnology or IntegrationMirror Security is an NVIDIA Inception member. The company leverages NVIDIA accelerated computing (CUDA, cuBLAS, NVIDIA NeMo framework, TensorRT-LLM) for its Encrypted AI Inference platform. NVIDIA's AI inference capabilities underpin Mirror's FHE production deployments.
• IntelcoreTechnology or IntegrationCollaboration to strengthen autonomous AI agents with fully homomorphic encryption. Intel's confidential computing (Intel Tiber Trust Authority) serves as the foundation for trustworthy AI deployments. Intel is developing purpose-built FHE silicon in collaboration with Mirror Security.
• MongoDBcoreTechnology or IntegrationPartnership to integrate Mirror Security's VectaX FHE engine with MongoDB's vector database capabilities for enterprise AI deployments. Enables encrypted vector search and memory for AI systems on MongoDB.
• AccopscoreTechnology or IntegrationPartnership integrating Accops HySecure (ZTNA service) with Mirror VectaX for zero-code-exposure AI coding. The combination enables enterprises to use AI coding productivity tools while maintaining IP privacy and compliance through ZTNA. Announced at Black Hat 2025.
• OWASP GenAIminorGTM or Marketing PartnerMirror Security participates in the OWASP GenAI initiative, aligning its products (DiscoveR) with OWASP Top 10 for LLMs and contributing to AI security standards. Listed as a partner on the OWASP GenAI initiative page.
• AAIF (Asia Africa Innovation Forum)minorGTM or Marketing PartnerListed as a partner on Mirror Security's website. AAIF is an innovation forum focused on Asia and Africa. Partnership likely provides conference participation and regional ecosystem access.
• SiSys AIcoreStrategic or Co-development PartnerStrategic collaboration announced August 2025 to develop next-generation Fully Homomorphic Encryption Co-Processor. The MIRROR Architecture aims to accelerate privacy-preserving computation beyond software limits, moving FHE operations to purpose-built silicon.

Scale indicators5 records

Recent moves7 records

Expansion highlights5 records

Direct peers

• Lakera: Lakera provides AI agent and LLM security including prompt injection defense, guardrails, and runtime protection. Its product scope overlaps directly with Mirror's AgentIQ and Zero offerings for AI agent governance.
• HiddenLayer: HiddenLayer provides AI security solutions including model scanning, adversarial defense, and runtime monitoring for ML systems. It competes directly in the AI threat detection and red-teaming category alongside Mirror's DiscoveR.
• Zama: Zama builds open-source FHE tooling (Concrete, TFHE-rs) for AI and blockchain applications. While it operates more as a developer platform than an enterprise product, its FHE-for-AI focus overlaps directly with Mirror's technology stack.
• Protect AI: Protect AI offers a comprehensive AI/ML security platform covering model scanning, red teaming, and runtime defense. Its product breadth across the AI security lifecycle maps closely to Mirror's DiscoveR, AgentIQ, and Zero.
• Enveil: Enveil provides FHE-based secure search and AI/analytics that compute on encrypted data. It serves financial services, healthcare, and government with privacy-preserving capabilities directly comparable to Mirror's VectaX engine.
• Duality Technologies: Duality Technologies offers an FHE-based secure data collaboration platform for AI and analytics. Like Mirror, its core technology is homomorphic encryption applied to AI/ML workloads, addressing regulated industries such as financial services and healthcare.
• Private AI: Private AI offers privacy-preserving ML tools including MPC and FHE-based redaction and inference for enterprise customers. Its focus on running models on sensitive data without exposure is functionally adjacent to Mirror's VectaX.
• Inpher: Inpher provides privacy-preserving machine learning and analytics using FHE and secure multiparty computation. Its enterprise platform targets financial services and government use cases parallel to Mirror's regulated-vertical positioning.

Broad incumbents

• Robust Intelligence (Cisco): Robust Intelligence was acquired by Cisco and is now part of Cisco AI Defense, offering AI firewall, model vulnerability scanning, and runtime guardrails. It represents a well-capitalized incumbent competing against standalone platforms like Mirror.

Emerging players

• Niobium Microsystems: Niobium develops purpose-built FHE hardware accelerators, addressing the same performance bottleneck Mirror is tackling with its SiSys AI co-processor partnership. It represents a complementary rather than directly competing approach to FHE acceleration.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/mirror_Security

Compliance3 records

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds1 record

Investors2 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Mirror Security