AI security
Start with an AI Exposure Review: what the workflow can access, which risks that creates, and what to fix first. Deeper testing is a separate written scope.
Controlled demonstration
Two fictional companies share one assistant. Run a cross-tenant retrieval, then turn on tenant isolation and run the same test again.
Open the playgroundMethod
No system access is required to start a conversation. Testing begins only after scope and authorization are agreed in writing.
You describe the AI system, its intended use, the data it handles, the tools it connects to, and the security concern in front of you. The first message does not ask for passwords, secrets, production credentials, patient data, or confidential datasets.
Before any testing or system access, both sides agree in writing on the scope, the objectives, the exclusions, the authorization, the testing boundaries, and the rules of engagement. Testing starts only after that explicit authorization.
Inside the agreed boundary, we test the attack surfaces that apply to the system.
You receive findings you can act on. Turnaround, a guaranteed outcome, and retesting are whatever the written engagement says. They are not assumed.
An introductory discussion can start with this, and nothing more sensitive.
Technical detail, evidence and any access are agreed after this first qualification. Do not send them in the public form.
Why now
Organizations are deploying AI into real work. Each system is a new path into data, tools and decisions.
What is being deployed
What that introduces
AI security cannot be added after deployment. It has to be engineered from the beginning.
Relevant to banks, insurers, healthcare, telecom, public administration, industry, universities and companies putting generative AI into production.
Secure Your AI01
Map AI systems, data flows, permissions, integrations and attack surfaces.
02
Test the system against the attacks that talk a model into doing the wrong thing.
03
Retrieval is a data path. If it can be poisoned or over-shared, the model will repeat it.
04
An agent with tools is a user that can act. We test what it is allowed to do.
05
A policy matters when it changes who can use a model, on which data, and with what oversight.
06
Models, tools and data change. The assessment has to change with them.
AI Security as a Service
Security testing for AI applications, chatbots, RAG systems, and AI agents. Identify vulnerabilities, understand the risks, and give your engineering team actionable remediation guidance.
01
3,500 MAD
A focused first assessment for teams that want an initial view of their AI application's main security risks.
Included
This is a focused assessment, not a full red-team engagement or security certification.
Request a Snapshot02
Recommended
7,500 MAD
A structured security assessment for software companies and AI startups deploying AI-powered products.
Included
Final scope depends on the application's architecture, integrations, number of workflows, and agreed testing boundaries.
Request an Assessment03
6,000 MAD/month
Recurring AI security testing for teams that continuously develop and release AI features.
Included
This is recurring, scoped security testing—not unlimited testing, 24/7 monitoring, or a guarantee of security.
Discuss a Monthly RetainerPrices in MAD; applicable taxes confirmed during quotation.
Additional applications, environments, integrations, and complex requirements may need a custom quote.
Testing begins only after scope and written authorization are agreed.
Attack surface
An AI system reads, decides and sometimes acts. Treat it as a privileged component of the infrastructure.
Your AI has access. Attackers want that access.
Request a Security Assessment01
User
Shadow AI, and prompts that leave with the person who typed them.
02
Application
Insecure integrations between the product and the model.
03Insider
AI model
Prompt injection, jailbreaks, system-prompt exposure, model manipulation.
04
Tools / agents
Excessive agency, tool abuse, malicious tool instructions, agent hijacking.
05
Data
Leakage, RAG poisoning, document injection, retrieval manipulation.
06
Infrastructure
Excessive permissions and weak boundaries around the system.
AI Exposure Review
A bounded assessment of what the workflow can reach, which risks that creates, and what to fix first. It is not a score, and it is not a certification.
FAQ
Direct answers. No invented client counts.
LINK TECHREATIONS. A startup in Casablanca, founded by Nour Kabbouri and Alaa Hamdoune, that secures AI systems, data, and digital products for organizations in Morocco. See AI security
A Moroccan cybersecurity company founded by Nour Kabbouri and Alaa Hamdoune. We secure AI systems, we audit and penetration-test infrastructure, and we build products. Medilink is in production. About the company
Yes. Applications, cloud, identity and infrastructure, plus AI red teaming when the system includes a model. See cybersecurity services
Yes. We assess models, RAG systems and agents: what they can see, what they can be told, what they can do, and what you can prove afterwards. See AI security
Our healthcare data-exchange product, in production, with encryption, traceability and privacy designed in. It is proof of how we build. Read the Medilink case study
Casablanca, Morocco. The work starts here. International work follows the same standard. Contact us
Use the contact form and a work email. Say what you are trying to secure, and whether an AI system is already deployed. Request an audit
No certification is claimed on this website. Practices are informed by OWASP guidance for applications and LLM systems, NIST CSF, MITRE ATT&CK, ISO 27001 principles, Law 09-08 and GDPR where it applies. See the method
No system access is required to start. This form is delivered to nourkabbouri@protonmail.com. Nour Kabbouri reads it. The founders lead the work that follows.
