👋 Year End Sale!

Day(s)

:

Hour(s)

:

Minute(s)

:

Second(s)

Buy Now
Study Later
You can buy a course now and start it whenever you want. It could be in a week, a month, or even a year. You can start your course when you're ready.
Practical DevSecOps - Hands-on DevSecOps Certification and Training.

In this blog

Share article:
Tells Google to show you more from Practical DevSecOps in AI, Search, and Discover.

AI Security Readiness in Saudi Arabia: The Complete Guide to the NCA Draft AI Cybersecurity Guidelines (AICG-1:2026)

Sneha Mukherjee
Sneha Mukherjee
Tells Google to show you more from Practical DevSecOps in AI, Search, and Discover.

Summary

A practical guide to AI security readiness in Saudi Arabia: the NCA draft AI Cybersecurity Guidelines (AICG-1:2026), breach costs in the Kingdom, the ten guidelines that need new engineering, control mapping, a 20-point checklist and a 90-day plan.

Saudi organisations are putting generative AI and agents into customer service, software delivery and operations. Security has not kept pace. In July 2026, the National Cybersecurity Authority (NCA) published a draft of what it expects: the AI Cybersecurity Guidelines (AICG-1:2026), with 42 guidelines across four domains.

This guide turns that draft into something you can act on: the threat picture, the business case, the structure of the guidelines, the controls and evidence behind each one, a self-assessment and a 90-day plan you can start on Monday.

Key takeaways

#TakeawayWhat it means for you
1The risk is already priced inThe average breach in the Kingdom costs SAR 27M. Public-facing applications, including AI chat and APIs, are the costliest way in at SAR 34.7M.
2October 2026 changed the clockPHP2Shell was exploited about five hours after its details leaked and over six hours before the fix shipped.
3The expectation is written downThe NCA draft sets 42 guidelines across 4 domains, names generative and agentic AI in scope and reuses the ECC structure.
4DevSecOps is the cheapest fixIBM links a DevSecOps approach, EDR and encryption to the lowest breach costs in Saudi Arabia.

Who this guide is for

•       CISOs and security leaders who need a defensible AI security roadmap and board-level numbers.

•       AppSec, DevSecOps and cloud security engineers who will build the controls.

•       AI and platform teams shipping LLM features, RAG systems or agents.

•       GRC and compliance teams mapping ECC, NCNICC and PDPL obligations to AI.

What’s inside

1.      The October 2026 vulnerability storm

2.      What the numbers say about AI and vulnerabilities

3.      What a breach costs in Saudi Arabia

4.      AI in the attack, and the defence

5.      The regulatory picture and the structure of AICG-1:2026

6.      Anatomy of an AI system: seven components, seven ways in

7.      Ten guidelines that need new engineering, and control mapping

8.      The 20-point readiness checklist

9.      Scoring and the 90-day plan, common mistakes, and how we help

10.   FAQ

1. October 2026: the vulnerability storm

One month taught one lesson: attackers now move in hours. In 31 days, defenders faced a record Patch Tuesday, firewall zero-days, a PHP flaw exploited before its fix, and a wave of bugs in AI agents and gateways.

PHP2Shell, hour by hour

On 21 October, two memory bugs in PHP itself (CVE-2026-109618 and CVE-2026-109619, CVSS 9.2) exposed every server running PHP 5.4 or later to unauthenticated remote code execution.

Three things stand out:

•       Exploitation started about five hours after the leak, and more than six hours before a fix existed.

•       The fastest defence was a WAF rule, live 33 minutes after the leak. Patching took over eleven hours.

•       The impact was national. Roughly 70% of websites run PHP, and France and the Netherlands took government sites offline to patch.

The rest of the month

DateIncidentWhat happenedReference
2 OctAWS Loom agent platformFull admin to any network client when no identity provider was setCVE-2026-103956
13 OctMicrosoft Patch TuesdayRecord 1,411 flaws; Windows Shell MotW bypass exploited as a zero-dayCVE-2026-96951
16 OctClaude Agent SDKCrafted session ID injects an MCP server and runs commands past permissionsCVE-2026-96620
19 OctEnvoy AI Gateway and Agent MeshAny OIDC token holder could self-enrol and call every MCP toolCVE-2026-103091
21 OctLiteLLM and agentic intrusionsMalware slipped into LiteLLM release packages; CrowdStrike reports four actors using AI agents in intrusionsSupply chain
21 OctPHP2ShellUnauthenticated RCE in PHP, exploited before the patchCVSS 9.2
28 OctCisco ASA and FTD firewallsZero-day exploited to crash SSL VPN devicesCVE-2026-76418
30 OctTAKO AI Agent for OktaUnauthenticated first-run admin takeover of an AI agentCVSS 9.8

Five of those eight incidents hit AI software directly: agent platforms, agent SDKs, AI gateways and an AI library’s release packages.

2. AI found the bugs. Attackers moved faster.

Disclosures have doubled, and AI-credited finds are up more than 20 times. Exploitation grew far more slowly. What really changed is speed, and AI software itself has become a target.

Metric (1 Jan to 23 Oct)2025 to 2026 change
CVEs flagged as exploited (CISA)1.2×
All CVEs published2.2×, reaching 86,292 (all of 2025: 48,151)
Critical CVEs2.9×
CVEs in AI software6.3×
CVEs credited to AI finders22.6×, from 34 to 769
Share of flaws leading to remote code execution50% for likely AI-found flaws vs 26% for others

What it means for an AI estate in the Kingdom

PriorityDraft guidelineAction
Know what you runAICG 2-13,191 CVEs hit AI software this year, and half sit in agent orchestration tools. Build an AI inventory first.
Patch at machine speedAICG 2-6-3Keep a kill switch for vulnerable AI features and a WAF rule path that ships in minutes.
Guard the gatewayAICG 2-2An AI gateway holds every provider key behind it. Vault the keys, scope the tokens, rate-limit.

3. What a breach costs in Saudi Arabia

IBM and the Ponemon Institute studied 602 breached organisations worldwide, including Saudi Arabia, for incidents between March 2025 and February 2026 (Cost of a Data Breach Report 2026).

Key figures

•       SAR 27M average cost per breach across all sectors.

•       SAR 38M in financial services, the most expensive sector, followed by industrial (SAR 35.4M) and technology (SAR 33M).

•       SAR 34.7M when the attack starts at a public-facing application, the costliest vector. AI chat interfaces and APIs fall into this category.

•       226 days to identify, 59 days to contain: a breach lifecycle of 285 days.

•       SAR 10M penalty for slowness: breaches with a lifecycle under 200 days averaged SAR 22M; those over 200 days, SAR 32M.

What moves the cost

Cuts the costRaises the cost
1. A DevSecOps approach1. Mismanaged secrets and keys
2. Endpoint detection and response2. Missing security skills
3. Encryption of sensitive data3. Poor threat prioritisation

Only 37% of breached Saudi organisations encrypted sensitive data at rest and in transit. That’s one of the cheapest gaps to close.

4. AI in the attack, and the defence

Attackers use AI to move faster. Defenders who use AI and automation well spend far less per breach. And every AI system you deploy is a new asset to protect.

AI is already in Saudi breaches

•       25% of malicious breaches in Saudi Arabia used AI. Across the Middle East the figure is 26%, with another 11% unsure.

•       55% of regional organisations with a SOC already run AI agents inside it.

•       16% of regional breaches began in the supply chain, at USD 8.45M each.

The automation dividend

Use of security AI and automationAverage breach cost in the Kingdom
Extensive useSAR 23.15M
No useSAR 32.08M
DifferenceSAR 8.9M saved

Yet 25% of Saudi organisations use no security AI or automation at all.

Why this matters for AI systems specifically

Agents run on service accounts, API keys and tokens. The region’s number one cost driver, mismanaged secrets and keys, is exactly the control agents depend on.

Teams in the region secure these non-human identities mainly through:

•       A machine identity inventory (57% of organisations).

•       Extending zero trust to AI-driven processes (43%).

In practice, that means every agent needs:

•       Its own identity, never a shared human account.

•       Least-privilege permissions scoped to the tools it actually uses.

•       Short-lived tokens and secrets held in a vault, never in prompts, code or config files.

•       Logging of every tool call, so the SOC can see what the agent did and on whose behalf.

5. The regulatory picture

The AI guidelines apply requirements the NCA already sets elsewhere to AI systems. A working ECC programme can be extended rather than rebuilt.

The road to AI regulation

WhenMilestoneWhat it means
2024ECC-2:2024108 essential controls: your baseline
2025NCNICC-1:2025Controls for the non-CNI private sector
Early 2026PDPL enforcementSDAIA enforces the personal data protection law
5 Jul 2026AICG-1:2026 draftAI guidelines open for public consultation
5 Aug 2026Consultation closesThe draft moves to finalisation
NextFinal guidelinesPublication pending as of November 2026

How the frameworks stack

Each layer builds on the one below it:

1.      PDPL at the base: personal data obligations underneath everything.

2.      ECC-2:2024 and NCNICC-1:2025 in the middle: the 108 baseline controls.

3.      AICG-1:2026 on top: AI-specific guidance.

Why act on a draft?

The draft calls the guidelines recommended, not mandatory. Four reasons to start anyway:

•       It restates NCA requirements you may already be measured on.

•       PDPL enforcement already targets weak technical security in AI data flows.

•       Clients and regulators will ask AI security questions shaped by this text.

•       Controls built now become evidence later.

Inside AICG-1:2026: 42 guidelines on one page

DomainGuidelinesSubdomains
1. Governance13Risk management · IT project management · Human resources · Awareness and training
2. Defense24Asset management · Identity and access · Network security · Data protection · Backup and recovery · Vulnerability management · Penetration testing · Logs and monitoring · Web application security
3. Resilience1Business continuity
4. Third party4Third-party cybersecurity
Total4215 subdomains

The draft covers the whole AI lifecycle:

•       Design and development: data, training, prompts.

•       Deployment: approval, guardrails.

•       Operation: monitoring, testing, patching.

•       Retirement: secure disposal and exit.

6. Anatomy of an AI system: seven components, seven ways in

Four components make a generative AI system. Agents add three more that let it plan and act. Each component maps to specific draft guidelines.

ComponentWhat it doesMain threatsDraft guidelines
AI modelLLMs and multimodal modelsExtraction, evasion, poisoning2-2-3, 2-4-1
Data and knowledge storesDocuments, databases, knowledge basesData leakage, poisoning2-4-1, 2-4-2
Retrieval and context (RAG)Fetches context for the modelIndirect prompt injection1-2-2
Application layerChat UIs, assistants, APIsPrompt injection, DoS, abuse2-4-4, 2-9-1
Planning (agentic)Breaks goals into tasksGoal hijack, runaway actions1-1-2, 1-2-7
Memory (agentic)Keeps context and resultsMemory poisoning2-4-3
Tool integration (MCP, APIs) (agentic)Connects agents to systemsTool abuse, excessive agency2-2-1, 2-8-1

October 2026, mapped to the components

DateAI agent CVEWhat happenedComponent
8 OctPydantic AI web chatAny website the developer visits can run the agent and its approval-gated toolsApplication layer
16 OctClaude Agent SDKCrafted session ID injects an MCP server and bypasses permissionsTool integration
19 OctEnvoy AI GatewayAny OIDC token holder can call every MCP toolIdentity and tools
30 OctTAKO AI Agent for OktaFirst-run admin takeover, CVSS 9.8Application layer

The agent risk in one line

Data from tools and memory can carry instructions. Treat it as untrusted, validate it, and log every tool call. October’s Claude Agent SDK, Envoy AI Gateway and TAKO agent CVEs all sit on this boundary.

What that looks like in practice:

•       Validate and sanitise everything entering the context window from tools, memory, RAG and third-party APIs.

•       Keep tool permissions per agent, not per platform, and require approval for high-impact actions.

•       Protect agent memory from unauthorised writes, and check stored content for poisoning.

•       Send every tool call and AI-initiated action to the SOC.

7. Ten guidelines that need new engineering

Most guidelines extend controls you already run. These ten need new designs, tests or pipeline gates. They take the longest, so start here.

#GuidelineDraft refWhat to doType of work
1Score agent actions by impact1-1-2Classify what each agent can do by impact, likelihood and reversibility; match safeguards to the scoreNew design
2Treat tool and memory data as untrusted1-2-2Validate anything entering the context window from tools, memory or third-party APIsNew design
3No uncontrolled learning in production1-2-3Don’t retrain live models unless the risk is assessed and controls are in placeNew pipeline gate
4Review AI-generated code1-2-5Security review, testing and approval before AI-written code reaches any environmentNew pipeline gate
5Graduated autonomy1-2-7Grow agent independence step by step, with human oversight of security decisionsNew design
6Rate limits against extraction2-2-3Usage thresholds and rate limits to reduce model extraction and data leakageNew pipeline gate
7Protect agent memory2-4-3Block unauthorised changes to memory and validate stored content against poisoningNew design
8Tested guardrails2-4-4Detect malicious input, unauthorised actions and disclosure; test before go-live and after changesNew test
9Red team and prove the override2-7-2Adversarial evaluation and red teaming, plus proof that human override worksNew test
10Safe shutdown and rollback3-1-1Rollback, safe shutdown and manual alternatives during an incidentNew test

From guideline to control and evidence

For each subdomain: the controls that satisfy it, the evidence an auditor will ask for, and the related OWASP Top 10 for LLM Applications (2025) risks.

D1: Cybersecurity governance

SubdomainWhat to implementEvidenceOWASP LLM
1-1 Risk managementAI risk register per use case; agent action catalogue scored by impact, likelihood and reversibility; threat model per AI systemRisk register, threat models, action matrixLLM06
1-2 IT project managementSecurity gates in the AI pipeline; context validation for RAG, tools and memory; AI-generated code scanned and reviewed; approval for high-impact AIPipeline configs, gate results, approvalsLLM01, LLM05
1-3 Human resourcesVetting for privileged AI access; acceptable use and confidentiality covering prompts and outputsSigned policies, vetting recordsLLM02
1-4 Awareness and trainingRole-based AI security training for builders; safe prompting awareness for all staffTraining plan, completion dataLLM02

D2: Cybersecurity defense

SubdomainWhat to implementEvidenceOWASP LLM
2-1 Asset managementInventory of models, datasets, vector stores, prompts, agents, MCP servers and AI vendors, with ownersAI asset inventory, AI-BOMLLM03
2-2 Identity and accessLeast-privilege agent roles; short-lived tokens; vaulted secrets; rate limits and quotas at the gatewayRole matrix, access reviews, gateway policyLLM06, LLM10
2-3 Network securitySegment AI components; egress allow-lists for model, tool and retrieval trafficNetwork diagrams, egress rulesLLM06
2-4 Data protectionEncrypt prompts, embeddings, weights, logs and outputs; classify data before it enters a model; memory integrity checks; input, output and action guardrailsEncryption settings, guardrail testsLLM02, LLM04, LLM08
2-5 Backup and recoveryVersioned backups of models, prompts, configs and vector stores; tested restore and rollbackBackup policy, restore logsLLM04
2-6 Vulnerability managementScan model files, libraries and containers; track AI advisories; kill switch for vulnerable featuresScan reports, advisory logLLM03
2-7 Penetration testingAI pen test before go-live and periodically; red teaming; human override drillsPen test and red team reportsLLM01, LLM07
2-8 Logs and monitoringLog prompt config changes, dataset access, tool calls and AI-initiated actions; AI detections in the SOCSIEM use cases, retentionLLM06
2-9 Web application securityTLS, rate limiting, DoS protection, controlled egress and unsafe-channel filters for AI apps and APIsWAF and API gateway configLLM10

D3 and D4: Resilience and third party

SubdomainWhat to implementEvidenceOWASP LLM
3-1 Business continuityRollback, safe shutdown and manual fallback per AI system; AI scenarios in BCM and DR testsRunbooks, DR test resultsLLM06
4-1 Third partyReview AI vendors, models and plugins before use and on a schedule; contract clauses for AI security, exit and disposalVendor assessments, contractsLLM03

8. The 20-point readiness checklist

A self-assessment you can finish in 15 minutes. Answer for your most important AI system: Yes = 2, Partly = 1, No = 0.

D1: Governance (6 questions)

#QuestionDraft ref
01AI security risks are assessed across the full lifecycle, from design to retirement.1-1-1
02Agent actions are classified by impact and reversibility, with safeguards for high-impact actions.1-1-2
03Data from tools, memory, RAG and third-party APIs is treated as untrusted and validated.1-2-2
04AI-generated code is security reviewed and tested before use.1-2-5
05High-impact AI needs a formal security approval before production.1-2-6
06Teams that build or operate AI get role-based AI security training.1-4-1

D2: Defense (10 questions)

#QuestionDraft ref
07We keep an inventory of all AI components with owners and lifecycle status.2-1-1/2
08AI agents and service accounts have least-privilege, role-based access.2-2-1
09Tokens, keys and AI service accounts are reviewed on a fixed schedule.2-2-2
10Rate limits and usage thresholds protect against extraction and data leakage.2-2-3
11AI components are segmented and egress is limited to approved destinations.2-3-1
12Prompts, embeddings, weights, logs and outputs are encrypted and access-restricted.2-4-1
13Guardrails are tested before release and after every material change.2-4-4
14AI systems are pen tested and red-teamed before go-live and periodically.2-7-1/2
15Prompt config changes, tool calls and AI-initiated actions are logged and monitored.2-8-1/2
16Vulnerable AI features can be disabled quickly until a fix is validated.2-6-3

D3: Resilience (2 questions)

#QuestionDraft ref
17Each AI system has a defined rollback, safe shutdown and manual fallback.3-1-1
18Backup, recovery and rollback for AI systems are tested periodically.2-5-2

D4: Third party (2 questions)

#QuestionDraft ref
19AI vendors, third-party models and plugins are assessed before use and monitored.4-1-2
20Contracts set AI security duties, exit, portability and secure disposal terms.4-1-3/4

Add up your total out of 40, then read your result in the next section.

Tips for an honest score

•       Score the system as it runs today, not as the design document describes it.

•       “Partly” means the control exists for some components or some of the time. If you can’t show evidence, score it “No”.

•       Ask the engineers who operate the system, not only the people who approved it.

9. Read your score, then run 90 days

What your score means

ScoreBandWhere to focus first
0 to 19ExposedStart with inventory, access and guardrails
20 to 31DevelopingClose gaps in testing, logging, agents and vendors
32 to 40AdvancedRed team, evidence and continuous assurance

Bands are a Practical DevSecOps heuristic for prioritising work, not an NCA rating.

The 90-day plan, week by week

PhaseWorkstreamWeeks
DiscoverAI asset inventory and AI-BOM1 to 2
 Agent action risk scoring1 to 3
 Threat model the top 3 use cases2 to 3
 Gap assessment against all 42 guidelines1 to 3
HardenLeast privilege and vaulted secrets4 to 6
 Input, output and action guardrails4 to 7
 Egress rules and AI logging to the SOC5 to 8
 AI code review gates in CI/CD6 to 8
AssureLLM and agent red team, then retest9 to 11
 Rollback and safe shutdown drill10
 AI clauses in vendor contracts9 to 12
 Board report and evidence pack11 to 12

How to adapt the plan to your band

•       Exposed: don’t skip Discover. Without an inventory, every later step guesses at scope. Prioritise vaulted secrets and least privilege for agents, since mismanaged keys are the region’s top cost driver.

•       Developing: compress Discover into two weeks and spend the extra time on guardrail testing and AI logging in the SOC.

•       Advanced: run Discover as a refresh, then move straight to red teaming, override drills and a repeatable evidence pack for regulators and clients.

Build it into the pipeline

IBM’s data links a DevSecOps approach to the lowest breach costs in the Kingdom. The same logic applies to AI. Controls such as AI code review gates, model file scanning, guardrail tests and egress policies are cheapest when they run automatically on every release, and every run produces evidence you can show an auditor.

10. Seven common mistakes to avoid

These are the patterns that most often slow AI security programmes down, or leave gaps that an auditor or attacker finds first.

#MistakeWhy it hurtsDo this instead
1Waiting for the final guidelinesControls take months to build, and October showed attacks land in hoursStart with the ten engineering guidelines now; they’re unlikely to disappear
2Treating AI as a separate programmeDuplicates governance and evidence you already produce for the ECCExtend your ECC controls, risk register and SIEM to AI components
3Shared keys for agentsMismanaged secrets are the region’s top breach cost driverOne identity per agent, vaulted secrets, short-lived tokens
4Trusting retrieved and tool dataIndirect prompt injection rides in through documents, email and APIsValidate everything entering the context window (AICG 1-2-2)
5Guardrails tested onceModel, prompt and tool changes silently break themRetest after every material change (AICG 2-4-4)
6No kill switchA vulnerable AI feature stays live while you wait for a vendorFeature flags and WAF rules that ship in minutes (AICG 2-6-3)
7Override never drilledNobody knows if a human can actually stop the agent mid-taskRun override and safe shutdown drills (AICG 2-7-2, 3-1-1)

11. How Practical DevSecOps can help

The Practical DevSecOps Consulting Team runs fixed-scope, fixed-timeline engagements that end with evidence you can show a regulator, a client or your board.

EngagementDurationWhat you get
AI Security Readiness Assessment (best starting point)3 to 4 weeksGap report against all 42 draft guidelines · AI asset inventory and risk register · Prioritised 90-day remediation plan
LLM and Agent Red Team2 to 3 weeksPrompt injection, tool abuse and MCP testing · Data leakage and model extraction checks · Human override validation and retest
Secure AI SDLC6 to 8 weeksGuardrails and AI code review in CI/CD · Agent identity, secrets and egress controls · AI detections and logging for the SOC
Agentic AI Threat Modelling2 weeks per use caseComponent and data flow mapping · Agent action risk scoring · Controls mapped to AICG and OWASP

How an engagement runs

1.      Scoping call (30 minutes): agree systems, goals and timeline.

2.      Assess: hands-on testing and review against AICG-1:2026.

3.      Fix with you: pair with your engineers on the priority gaps.

4.      Evidence pack: report, retest results and audit-ready evidence.

Bring your checklist score and your top AI use cases to a free 30-minute call, and leave with your gaps and the first steps to close them.

Build the skills before the final text lands

Every step in this guide depends on engineers who can threat model an AI system, run a prompt injection, abuse an over-permissioned agent, and build guardrails that hold.

The Certified AI Security Professional (CAISP) course from Practical DevSecOps teaches exactly that, in browser-based labs:

•       The OWASP Top 10 for LLM Applications, with hands-on prompt injection and data leakage labs.

•       MITRE ATLAS and STRIDE threat modelling for AI systems.

•       AI supply chain security: model scanning, signing, SLSA and MLBOMs.

•       AI governance under the NIST AI RMF, ISO/IEC 42001 and the EU AI Act.

•       No prior AI or ML experience needed. It ends with a 6-hour practical exam.

Start your 7-day free trial and run your first AI security lab today.

Certified AI Security Professional (CAISP)7-day free trial

Open a live AI security lab in your browser today

Real targets, real terminals, no local setup.

Start your free trial
No credit card required.

Frequently Asked Questions

What is AICG-1:2026?

 The NCA’s draft AI Cybersecurity Guidelines. It sets 42 guidelines across four domains (governance, defense, resilience and third party) and names generative and agentic AI in scope.

Are the AI Cybersecurity Guidelines mandatory?

The draft calls them recommended, not mandatory. They restate NCA requirements many organisations are already measured on, and PDPL enforcement already targets weak security in AI data flows.

When will the final guidelines be published?

Consultation ran from 5 July to 5 August 2026. As of November 2026, the final version is still pending.

How do the AI guidelines relate to the ECC?

They reuse the ECC structure and apply existing NCA requirements to AI systems. A working ECC-2:2024 programme can be extended rather than rebuilt.

Which guidelines should we start with?

The ten that need new engineering: agent action scoring, untrusted context, no uncontrolled learning, AI code review, graduated autonomy, rate limits, agent memory protection, tested guardrails, red teaming with override proof, and safe shutdown.

What is the biggest AI security risk for Saudi organisations right now?

 Agents with too much access. They run on keys and tokens, mismanaged secrets are the region’s top breach cost driver, and October 2026’s agent CVEs all sat on the boundary between agents and their tools.

Do our engineers need machine learning expertise?

No. Most of the work is familiar security engineering (least privilege, input validation, logging, supply chain and pipeline gates) applied to new components.

Sneha Mukherjee

Sneha Mukherjee

Security Research Writer

Sneha Mukherjee is a Content SEO Specialist specialising in AI security, cybersecurity, SEO content strategy, and technical content. She focuses on creating clear, research-driven content that helps businesses communicate complex AI and security topics effectively while improving search visibility and audience engagement.

Related articles

Start your journey today and upgrade your security career

Gain advanced security skills through our certification courses. Upskill today and get certified to become the top 1% of cybersecurity engineers in the industry.