Blue background

AI Security: We make your AI usage secure, controllable and auditable.

Your teams work with ChatGPT, Copilot, and custom agents, whether authorized or not. We provide visibility into which AI is in use, establish the necessary guardrails, and subsequently test whether they can withstand an attack.

Schedule an Initial Consultation

AI Is Adopted Faster in Companies Than the Rules Governing It

Artificial intelligence is part of everyday work and an integral component of operational processes in many companies. Assistance systems, AI-powered software development, and autonomous agents give rise to risks that established security and compliance frameworks are not designed to address: the leakage of confidential data to external services, the manipulation of models through tampered inputs, and the limited traceability of automated decisions.

The EU AI Act defines binding requirements for transparency, AI competence, and risk management in this regard. ISO/IEC 42001 describes the structure of an AI management system, while the BSI AIC4 criteria catalog specifies security requirements for AI cloud services.

Warning sign with bomb icon

Shadow AI

Use of unauthorized AI services with corporate data.

Encrypted message icon

Prompt Injection

Manipulation and control of AI agents via crafted inputs.

Shield with checkmark icon

Verification Requirements

Transparency obligations since 08/2026, high-risk AI from 12/2027.

Status: 09/2026

The use of generative applications is now well established in small and medium-sized businesses. Every integration and every agent increases the attack surface. Customers and auditors are increasingly demanding evidence of controlled use.

57 %

Companies with 20 or more employees use AI (Bitkom, 09/2026; 2024: 20%)

78x

Growth in bot traffic in 2025 (HUMAN Security Benchmark Report 2026, September 2026; 2026)

25%

All security incidents in 2026 are AI-powered or automated (Deepstrike AI Cyber Attack Statistics 2026, September 2026; August 2026)

Portrait of Timm Börgers, Managing Director at carmasec.
The pressure to quickly put generative AI applications into productive use is justified. Speed and security are not mutually exclusive. What is crucial is that governance, technical safeguards, and continuous testing work together from the very beginning. Any vulnerability uncovered during testing must be addressed in policies and configurations. Only then will cybersecurity be effective in the long term.

Timm Börgers, Managing Partner & Trusted Advisor

Request a free consultation now

How Best to Get Started

Every AI Security service aligns with one of our three pillars. Information Security & Compliance sets the direction, Defense Security implements it, and Offensive Security verifies its resilience.

ISM & Compliance

AI Security Posture Check

A comprehensive baseline assessment as the starting point for any further action. Structured interviews and, upon request, an evaluation of your proxy or DNS logs reveal which services are in use, including Shadow AI. The result: a maturity traffic light assessment for each action area and a prioritized roadmap.

Request Posture Check

ISM & Compliance

AI Governance & AI Act Readiness

AI inventory, AI policy, roles, and gap analysis based on ISO/IEC 42001, seamlessly integrated into your existing ISMS.

 

Defense Security | Securing AI Usage

Secure AI Rollout

Securely introduce Microsoft 365 Copilot, ChatGPT Enterprise, and other assistants: clean up permissions, classify data, configure DLP, harden the tenant, and thoroughly test from an attacker’s perspective.

Learn More

Defense Security | Securing AI Usage

Secure AI Access Layer “caia”

A centralized layer between your users and all large language models: roles and permissions, sensitive data filtering, guardrails, and audit logging for every prompt. Built within your infrastructure and handed over to your team.

Yes, sounds right

Offensive Security | Testing and Validation

Pentesting of AI/LLM Systems

Testing chatbots, assistants, and RAG applications: prompt injection, jailbreaks, data exfiltration, privilege escalation.

Learn More Now

Offensive Security | Testing and Validation

AI Agent & MCP Security Assessment

Testing AI agents and MCP servers: Can the agent be redirected? Does it abuse tools? Does it hold excessive permissions?

 

Offensive Security | Testing and Validation

AI-Augmented Red Teaming & Adversary Simulation

Objective-driven adversary simulation over several weeks, enhanced by AI tools. Evaluates whether your SOC detects attacks executed with modern AI capabilities.

 

Offensive Security | Testing and Validation

vincora AI Agent

Our proprietary AI agent for offensive testing. vincora operates within a strictly defined technical scope. Our operators oversee every step and manually authorize each vulnerability exploitation.

Inquire Now

Auditing AI Agents and MCP Servers

Agents read emails, create tickets, and access files, databases, and APIs via the Model Context Protocol (MCP). We evaluate whether your agents can be hijacked, whether they abuse integrated tools, and what permissions their identities actually hold—aligned with the OWASP Top 10 for Agentic Applications and the OWASP MCP Top 10.

View Agent Assessment
PROCESS

Your Path to Resilient AI Security

Step 1: Security posture analysis

Understand

Which AI tools are you using, what data flows where, and who authorized it?

  • Maturity traffic light assessment and roadmap from the AI Security Posture Check.
Step 2: Definition of measures

Govern

AI inventory, policy, roles, and approval workflows integrated into your ISMS.

  • Actionable AI governance applied in daily operations (AI Governance & AI Act Readiness).
Step 3 in the security process

Secure

Centrally govern access to language models and harden commercial AI tools.

  • Gateway and hardened tools deployed in your environment (AI Governance Gateway, Secure AI Rollout).
Step 4 in the security process

Validate

Test applications, agents, and SOC detection capabilities under real-world attack conditions.

  • Findings complete with attack vectors and remediation measures (LLM Application Pentest, Agent & MCP Assessment, AI-Augmented Red Teaming).

Why carmasec Is the Right Partner for AI Security

We combine years of experience in information security, defense, and offensive security with practical knowledge of generative systems, and we treat AI with the same care as any other business-critical infrastructure.

As soon as your teams start using AI, we ensure that data, rights, and decisions remain under your control.

  • Security Expertise Meets New Technology
    Since 2018, we have been managing ISMS projects, penetration tests, and defense architectures for organizations ranging from small and medium-sized businesses to large corporations. We apply this methodology to generative applications, drawing on our own practical experience: AI is an integral part of our penetration tests and red-teaming projects.
  • Your Infrastructure, Not a One-Size-Fits-All Solution
    Our recommendations are tailored to your infrastructure. We integrate solutions you’re already using, build on open source, or implement solutions from our technology partners—and we disclose every partnership in our proposal.
  • A Team
    Peer-to-peer from the very first conversation: The consultants who draft your AI policy work alongside the engineers building the gateway and the penetration testers attacking it. Findings from the test are directly incorporated into the policy and configuration.
  • Plain Language Instead of Hype
    We clearly tell you which measures your current maturity level requires and which can wait. From individual building blocks to a complete AI management system: You invest where it makes a difference.
  • AI in Our Own Practice
    We use AI ourselves in penetration tests and red-teaming projects, with vincora as our proprietary tool. That’s why we know from hands-on experience where agents excel and where they need oversight.

 

Clarify the starting point in 30 minutes

Smiling portrait of Niklas Ereth, Senior Security Consultant at carmasec.
AI is accelerating every change in IT. Governance guidelines that take effect before a prototype goes live keep pace with this momentum.

Niklas Ereth, Senior Security Consultant

Frequently Asked Questions About AI Security

Where should we begin?

If you are unsure which AI tools are currently in use across your organization, start with the AI Security Posture Check. If you have an active project underway, such as a Copilot rollout or a proprietary chatbot, you can start directly with the targeted service. We can clarify this in a 30-minute initial consultation.

Do we need to comply with the EU AI Act already?

Partially. Prohibitions on certain AI practices took effect on February 2, 2025; obligations for providers of general-purpose AI models have applied since August 2, 2025; and transparency obligations (e.g., for chatbots) apply from August 2, 2026. Organizations must also implement measures to ensure AI literacy among their workforce. Under the Digital Omnibus, requirements for high-risk AI apply from December 2, 2027, or August 2, 2028 (as of 09/2026). We handle the technical classification of your systems. Legal assessments remain the responsibility of your legal department or external counsel.

Does carmasec sell its own proprietary AI security product?

We do not sell a proprietary AI security product. We develop solutions based on open source, integrate platforms you already utilize, or deploy solutions from our technology partners. Any vendor partnerships are fully disclosed in our proposals. Vincora is our internal tool for offensive testing and is not for sale.

Do you use third-party AI models during assessments?

We transparently disclose in our proposal which models we utilize and where they are hosted. Upon request, we work exclusively with models hosted within Europe.

What size organization are these services designed for?

For mid-sized enterprises and large corporations. The Posture Check and Secure AI Rollout are tailored specifically for mid-sized organizations. Red Teaming is intended for organizations operating their own SOC.

Alongside our core services, we support you with:

Person on growth arrow icon
Selection of suitable models and tools from your existing inventory, open-source options, or technology partnerships
Person at strategy whiteboard icon
AI literacy training for your staff in accordance with the EU AI Act
Document with shield icon
Grant funding applications for digitization and AI initiatives
AI chip icon
Security assessments for AI agents and MCP servers

Contact

Ready for AI with guidelines?

Tell us which AI tools you’re using or planning to use. In our initial consultation, we’ll assess where you stand and determine the best next step.

  • Free initial consultation
  • 30 minutes
  • Specific Assessment
Portrait photo of Niklas Ereth at carmasec.
Niklas Ereth
Senior Security Consultant
+49 (0)201 426 385 905