Contact

AI

Cybersecurity area

AI

AI security protects generative AI, LLM applications, agents, inputs, outputs, data sources, permissions, and integrations into business processes.

Thematic hero image for the AI area

Acronym Glossary

This glossary explains acronyms used in this part so the intended security function is clear.

AI
Artificial IntelligenceArtificial intelligence used for analysis, content generation, decision automation, or process support.
PII
Personally Identifiable InformationPersonal data that can directly or indirectly identify a specific person.
RBAC
Role-Based Access ControlAccess control based on roles where permissions follow the user's job function or responsibility.
RAG
Retrieval-Augmented GenerationAn approach where an AI answer is grounded in retrieved sources or documents rather than model memory alone.
LLM
Large Language ModelA large language model used to process and generate text, answers, or natural-language interactions.
MCP
Model Context ProtocolA protocol for connecting AI tools and agents to external data, systems, and actions.

What This Area Covers

AI security addresses risks related to generative AI, large language models, RAG systems, agents, plugins, connectors, and access to internal data. It connects application security, data protection, identity, governance, AI usage monitoring, and control of automated actions.

Why It Matters

AI systems may process sensitive data, interpret external content, call tools, and generate responses that affect decisions or processes. Risks include prompt injection, leakage of personal data, bypass of permissions, unverified outputs, weak source control, excessive agentic autonomy, and misuse of integration layers such as MCP servers.

How To Navigate This Area

Generative AI introduces new risks: prompt injection, sensitive data leakage, unverified outputs, excessive agent permissions, weak separation of sources, and new integration interfaces. Practical orientation must combine data protection, application security, identity, governance, and monitoring of AI usage including RAG and MCP scenarios.

AI security technologies should be separated into input protection, output filtering, response validation, RBAC filtering, RAG trust, and security of integration layers. Many organizations first need basic visibility into where AI is used, what data it touches, and which actions it can perform before they implement detailed technical policies.

The technology list below works as a map of the AI area. The practical sequence is to name the security problem first, select the closest technology category second, and only then compare relevant vendors. This helps separate similar-sounding solutions and reduces the risk of adding another tool without a clear architectural role.

Technical Section Schema

Technical schema for the AI area
The schema shows the main inputs, security layers, and outputs of the AI area.

Questions To Ask When Selecting A Solution

  • Which AI tools are used internally and which are integrated into processes?
  • Does AI work with sensitive data, personal data, or internal documents?
  • How are inputs, outputs, sources, and user permissions controlled?
  • Should AI only answer, or can it call tools, change data, and execute actions?
  • Do AI agents or MCP servers access systems that require audit and restrictions?

How To Use This Section

Each technology page expands one category, explains its architectural role, and lists SecTec portfolio vendors related to that area. The section page is therefore the orientation layer, while the detail pages support concrete mapping to solutions.

Vendors From The SecTec Portfolio

Attack coverage and defensive context

MITRE ATT&CK® Enterprise version 19.1 is used as a reference model. The mapping is a SecTec expert interpretation and does not imply endorsement by MITRE.

Practical attack scenarios

  • Phishing and identity abuseAn adversary manipulates a user, obtains credentials, or abuses a trusted account to gain access to the environment.
  • Public application or API abuseAn adversary exploits a vulnerable web application, API, client-side code, or software component for initial access and execution.
  • Cloud and SaaS compromiseAn adversary abuses cloud identities, permissions, configurations, or application integrations to access services and data.
  • Malware and Command and ControlMalicious code executes, maintains access, and communicates with adversary infrastructure through common or covert channels.
  • Data exfiltration and leakageAn adversary searches for, collects, and transfers sensitive data from endpoints, cloud services, email, or internal repositories.
  • Supply chain and trusted accessAn adversary abuses a supplier, software dependency, external access path, or trusted relationship to bypass normal controls.
  • Business Email Compromise and mailbox takeoverAn adversary takes over or impersonates a trusted email identity and abuses communications for fraud, payment diversion, data collection, or further phishing.
  • Privilege escalation and persistenceAn adversary expands privileges, creates or modifies accounts and policies, and establishes repeatable access to systems, identities, or cloud environments.
  • Insider threat and authorized access abuseAn insider or an adversary with a valid account abuses legitimate permissions to discover, copy, manipulate, or transfer sensitive data.
  • Injection and command executionUntrusted input changes the meaning of a command, query, or template and enables data access, logic bypass, or code execution.

Relevant ATT&CK techniques

Facebook Twitter Linkedin Youtube
Contact