The AI, automation and cybersecurity terms we use every day, explained in a sentence or two.
The EU regulation on artificial intelligence (EU 2024/1689). It classifies AI systems by risk level and assigns proportionate obligations to those who build and use them.
Software that uses a language model to understand a request and complete it with real tools, such as a calendar, CRM or business system, within set permissions and rules.
Guide: AI agents for business Service: AI agents and voice assistants
The skills that let people use artificial intelligence knowingly, aware of its limits and risks. The AI Act asks those who provide or use AI systems to promote it.
A system that answers calls in natural language, understands the request, acts on the company's systems and passes to a person the cases that need one.
Guide: AI phone assistant Service: AI agents and voice assistants
An interface that lets two pieces of software exchange data and commands in a controlled way. It is the standard way to integrate business systems, CRMs and external services.
All the points from which an attacker can try to get into a system: exposed services, APIs, accounts, devices, suppliers.
Continuous integration and delivery: every code change is checked with automated tests and released only if the checks pass.
A unique identifier assigned to a publicly known security vulnerability, used worldwide to refer to the same flaw.
A standard scoring system, now at version 4.0, that rates vulnerability severity from 0 to 10 to decide what to fix first.
In the AI Act, whoever uses an AI system under their own authority for their own activity. It is the role of most companies that use AI.
A numeric representation of a text that captures its meaning. It allows documents to be searched by meaning rather than just by keywords.
Turning data into a form that only key holders can read. It protects data stored on servers and data travelling over the network.
The risk that an AI agent has more permissions, tools or autonomy than its task requires. It is in the OWASP Top 10 for LLM applications.
Extra training of a model on specific data to adapt its style or skills. It is often unnecessary: well-built instructions and context are enough.
Controls that limit what an AI system can say or do: input and output filters, deterministic rules, confirmations for risky actions.
A language model answer that sounds right but contains false or made-up information. It is reduced by checking quoted data against real sources.
Reducing the attack surface of servers, applications and devices: unused services off, secure configurations, updates and minimal access.
A cryptographic signature that verifies a message, such as a webhook or a link, really comes from the sender and has not been altered.
An approach where a person approves an automated system's important or uncertain decisions before they become actions.
The property whereby repeating an operation has no double effect: an order or payment is processed once even if it arrives twice.
Guide: 12 processes SMBs can automate with AI (and how to get started) Service: AI automation and integrations
A technique to get a language model to bypass its own safety rules, usually with cleverly worded requests.
The principle that every person, service or AI agent gets only the permissions essential to its task, and nothing more.
An AI model trained on large amounts of text to understand and generate language. Claude, GPT and Gemini are examples of LLMs.
An open protocol to connect AI models to external tools and data in a standard way, with authentication and permissions.
Guide: AI agents for business Service: AI agents and voice assistants
A lightweight protocol for connected devices and servers to communicate, widely used in the Internet of Things.
Sign-in that requires at least two different proofs, such as a password and a code on the phone. It blocks most logins with stolen credentials.
An architecture where a single software serves several companies while keeping each one's data separate, for example with a dedicated database schema.
A platform to build automation workflows between applications, which can also be installed on your own servers to keep data under control.
Guide: 12 processes SMBs can automate with AI (and how to get started) Service: AI automation and integrations
A device software update sent remotely over the network. It must be signed so the device only installs genuine versions.
A non-profit foundation that publishes open guides and standards on software security, including the web Top 10 and the Top 10 for LLM applications.
A simulated, authorised attack on an application, API or network to find out which vulnerabilities are really exploitable and how to fix them.
A deceptive message, by email, text or phone, that tricks people into revealing credentials or installing malware by posing as a trusted source.
The instructions and context sent to a language model. The system prompt holds the rules set by whoever builds the application.
An attack that slips malicious instructions into the text a language model reads, to make it ignore its rules or take unauthorised actions. It is the top risk in the OWASP Top 10 for LLMs.
A technique that retrieves relevant documents from a knowledge base and passes them to the model before it answers, so answers rest on company data.
Malicious software that encrypts a company's data and demands a ransom to restore access. Separate, tested backups are the first defence.
A deliberate attack on a system by specialists to find its weak points before a real attacker does. It is also used for chatbots and AI agents.
A database rule that shows each user only the rows that belong to them, even if the application gets a query wrong.
An approach where security is designed in from the start, with threat modelling, least privilege and automated checks, rather than added at the end.
The use of AI tools not approved by the company, often with personal accounts, which exposes confidential data outside any control.
Technology that turns speech into text in real time. It is the first step of every voice assistant.
Technology that turns text into a natural voice. In a phone assistant it gives voice to the model's answers.
A structured analysis of what can go wrong in a system, who might attack it and how, done before development to decide the defences.
A broad, largely automated analysis that finds and ranks the known vulnerabilities of systems and applications.
An automatic notification one service sends another when something happens, such as a payment or a new lead. It should be verified with a signature.
A security model where no user or device is trusted by default: every access is verified, even from inside the network.
How AI, automation and security really work, explained without jargon and with sources.
What an AI agent is, how it differs from a chatbot, which SMB processes it can handle and how to introduce it with permissions, controls and costs in check.
6 min read
What prompt injection is, why it tops the OWASP Top 10 for LLM Applications and which defences actually work for chatbots and AI agents.
5 min read
The EU AI Act explained for businesses: risk levels, obligations in force, chatbot transparency from 2 August 2026 and new deadlines after the Digital Omnibus.
5 min read