The New Attack Surface

Your models and agents are now the target.

As organizations move their operations onto AI, the AI stack becomes critical infrastructure — and critical infrastructure gets attacked. Prompt injection attacks, model poisoning, training data contamination, agent privilege escalation, and supply chain compromise through third-party model dependencies: these are not theoretical risks. They are active threat patterns with documented case studies.

Traditional cybersecurity frameworks were built for software systems with deterministic behavior. AI systems behave probabilistically, can be steered through natural language, and have complex supply chains that extend to foundation model providers. Securing them requires new frameworks, new tooling, and new threat models.

"Model governance, prompt injection defense, agent guardrails, pipeline integrity — because your AI stack is now critical infrastructure. We align with every KSA regulation that governs it."
Threat Modeling Frameworks

The frameworks that map your AI attack surface.

OrwyTech applies three complementary frameworks to build complete threat modeling coverage for any AI deployment. Each addresses a distinct scope — together they map the full attack surface from model-level threats through agentic deployment architectures.

MITRE ATLAS v4
Adversarial Tactics for AI — Nov 2025, Updated Feb 2026
The adversarial TTP taxonomy for AI systems, updated through 2026 to cover agentic architectures, multi-agent deployments, and tool-use abuse. Our engagements use ATLAS as the primary reference for AI threat modeling, red-teaming, and detection engineering.
OWASP LLM TOP 10
2025 Edition — The Definitive LLM Risk Register
The definitive risk register for LLM deployments — covering prompt injection, insecure output handling, training data poisoning, supply chain vulnerabilities, excessive agency, and model theft, among others. Every LLM deployment we deliver is audited against the full Top 10 before go-live.
MAESTRO (CSA, Feb 2025)
7-Layer Threat Modeling for Agentic AI
Cloud Security Alliance's layered framework for agentic AI threat modeling — spanning foundation models, data operations, agent frameworks, deployment infrastructure, observability, compliance, and the broader agent ecosystem. Our preferred reference for securing multi-agent and autonomous AI deployments.
Governance Standards

NIST, ISO, and the full compliance stack.

AI security isn't only about technical controls — it's about governance. We align every engagement with the recognized international standards and the KSA-specific regulations that govern AI deployments in Saudi Arabia.

NIST AI RMF
AI Risk Management Framework
GOVERN → MAP → MEASURE → MANAGE. The four-function framework for managing AI risk across the full lifecycle. Used as the governance backbone for every AI security engagement.
ISO/IEC 42001
Certifiable AI Management System Standard
The international standard for AI management systems — auditable, certifiable, and recognized by regulators worldwide. We align AI governance implementations with ISO 42001 requirements.
Secure AI Framework
Secure-by-Design AI Controls
Practical secure-by-design controls for the AI systems you build and operate. This framework closes the gap between AI deployment speed and security maturity — the implementation playbook behind our Cyber for AI engagements.
KSA Regulatory Landscape

Every regulation that governs your AI in Saudi Arabia.

Saudi Arabia has moved faster than most jurisdictions in establishing mandatory AI security requirements. For organizations operating in KSA — whether government entities, CNI operators, private sector AI companies, or financial institutions — compliance with these frameworks is not optional. We know them precisely and build to them by default.

SDAIA
AI Risk Classification Policy (Draft)
SDAIA's four-tier AI risk classification: Critical / High / Limited / Low. Each tier has mandatory requirements for security testing, continuous monitoring, and system registration. We assess your AI deployments against all four tiers and implement the required controls.
NCA ECC-2:2024
Essential Cybersecurity Controls — Government & CNI (Mandatory)
Mandatory for all government entities and critical national infrastructure operators. NCA ECC-2:2024 explicitly covers AI and MLOps environments — secure AI development practices, AI system monitoring, and AI risk management controls. Released 2024, full enforcement underway.
NCA NCNICC-1:2025
National Cybersecurity Controls for AI Companies in Private Sector (Mandatory)
Released January 2026. Mandatory for private sector AI companies operating in KSA. Covers AI system security requirements, data handling for AI, agent governance, and incident reporting for AI-related security events. This is the most directly relevant regulation for commercial AI deployments.
PDPL
Personal Data Protection Law — AI/Automated Decision Safeguards
PDPL extends to AI systems that process personal data or make automated decisions. Requirements: data minimization in training, meaningful consent for AI-based profiling, right to explanation for automated decisions, and breach notification when AI systems are involved.
SAMA CSF
Cybersecurity Framework — Financial Sector AI Models
SAMA's cybersecurity framework explicitly requires financial institutions to apply cybersecurity controls to AI models and automated decision systems. Credit scoring models, fraud detection systems, and robo-advisory tools all fall in scope.
MITRE ATLAS v4 OWASP LLM Top 10 MAESTRO Framework NIST AI RMF ISO 42001 Secure-by-Design AI NCA ECC-2:2024 NCA NCNICC-1:2025 SDAIA Compliance PDPL SAMA CSF
Secure Your AI

Your AI stack is critical infrastructure.
Treat it that way.

We assess your current AI deployments against MITRE ATLAS, OWASP LLM Top 10, MAESTRO, and the full KSA regulatory stack — then build the controls to close every gap.

Request an AI Security Assessment Next: Cloud Transformation →