Skip to main content
010 447 2445Contact
Arke Technologies home
Request a hardware quoteTalk to us
Arke Technologies home

How we use AI

Last updated: September 2026

Arke Technologies (Pty) Ltd, a Johannesburg technology company, uses AI in defined parts of its work under one rule: AI recommends and people decide. This AI use disclosure sets out where we use AI, which models we use, how we protect the data involved and what you can ask of us.

Ask about our AI useRead our privacy notice
Home

How we use AI

Last updated: September 2026, 14 min read

Arke Technologies (Pty) Ltd, a Johannesburg technology company, uses AI in defined parts of its work under one rule: AI recommends and people decide. This AI use disclosure sets out where we use AI, which models we use, how we protect the data involved and what you can ask of us.

Last updated: September 2026, 14 min read

On this page

What this AI use disclosure covers

This disclosure is published by Arke Technologies (Pty) Ltd, registration number 2017/386810/07, whose head office is at 8 Greenstone Place, Stoneridge Office Park, Block A North, First Floor, Greenstone Hill, Johannesburg, 1609, South Africa. It covers the whole company, including Server Hub, our IT hardware division, and Fibre Boyz, our connectivity division.

It describes three things: the AI we use to run our own business, the AI features in the platforms we own, and the approach we take when we build AI into systems for clients. By AI we mean software that generates, summarises, classifies or recommends, most often large language models, the technology behind AI chat assistants.

We publish it because transparency is one of our principles, and because the organisations we serve need it before they buy: boards applying King IV Principle 12 to technology and information governance, procurement officers evaluating a bid, and schools responsible for learner data. It sits alongside our privacy notice, website terms of use and cookie notice.

Our rule: AI recommends, people decide

We design and use AI under Emergent Intelligence, our framework for AI that supports human judgement rather than replacing it. Its five principles apply to our own operations and to every AI system we build for a client.

  • Agency preservation

    AI provides analysis and recommendations; people make the decisions that matter. We do not build "black box" automation that removes human judgement from a critical decision.

  • Transparency

    Every output should be explainable: what informed it, and how far to rely on it. You should know when you are dealing with AI.

  • Governance and auditability

    Every AI action is logged and reviewable, so an organisation can see what the AI did and who approved it.

  • Progressive enhancement

    We start with low-risk uses, check how the AI performs and expand only as confidence builds.

  • Ethical boundaries

    Clear limits on what AI may do, what data it may see and which decisions stay strictly human.

Where we use AI today

This table lists where AI runs in our business and our products, what it does there and the control that applies to it.

WhereWhat the AI doesThe control
Server Hub sourcing and quotingHelps our team find and compare products, check prices and availability, and prepare quotes for Server Hub, whose catalogue lists more than 26,000 productsEvery priced quote line is backed by a dated, verifiable source, and AI actions are logged and reviewable
Server Hub customer assistantAnswers customers' product and availability questions on serverhub.co.zaIt can be switched off, set to draft replies that staff approve before they are sent, or run automatically, and a customer can always ask to speak to a person
AtlasOS AI-assisted advisoryGives staff at professional services firms conversational access to their own practice data: overdue tasks, compliance risks and client summariesIt supports staff decisions without replacing their judgement, and each firm's data is isolated from every other firm's
Drafting and engineeringHelps our team draft and edit written material, including pages and guides on this website, and helps our engineers write and test codeA person reviews, edits and approves the output before it is published or shipped, and checks facts and figures against their sources
Systems we build for clientsAssistants, document processing and workflow automation designed for a client's own workThe client sets the approval points, following the steps described below

AtlasOS is our platform for professional services firms; visit the AtlasOS site  (opens atlaspractice.app) to see how firms use it. The Server Hub commerce and quoting platform case study shows the quoting and sourcing tools in more detail.

Our managed IT service also runs automation that acts on client systems, and we hold it to the same rule. Monitoring checks each computer every 15 minutes and logs faults as tickets automatically, and the riskier repairs in our standard library need a recorded human approval before they run; every remote session is approved and logged. We are building the same approval pattern into Hesper, our IT operations platform, which launches in November 2026 with early access until then. The IT service desk page explains how tickets reach our team.

What we do not do with AI

  • We do not make decisions about you based solely on automated processing that have legal consequences for you or affect you to a substantial degree, as section 71 of the Protection of Personal Information Act, 2013 (POPIA) requires.
  • We never train AI models on client data without the client's explicit consent.
  • We do not send a model more information than its task needs.
  • We do not present an AI assistant as a person.
  • We do not let riskier automated changes run on client systems without a recorded human approval.
  • We do not sell or rent personal information, to AI providers or to anyone else.

The AI models we use

We build with models from Anthropic (Claude) and Google (Gemini, through Google Cloud's Vertex AI), and with open-weight models, which an organisation can run on its own hardware. We choose the model for each job, weighing accuracy, cost and where the data may be processed, rather than tying every task to one provider.

Naming these providers describes the technology we use; it does not imply a partnership or an endorsement. When a model provider processes personal information for us, we treat it as a service provider under a written agreement that requires confidentiality and security safeguards, as our privacy notice describes, and any processing outside South Africa takes place only as section 72 of POPIA allows. Our Information Officer can tell you which provider handles a particular service.

For information that should not leave the country, we design private AI deployments that keep sensitive data in South Africa and run open-weight models where they fit, on servers and AI workstations supplied through Server Hub. See private AI and AI infrastructure.

How we protect data used with AI

The safeguards that section 19 of POPIA requires for personal information apply in full when AI is involved. In practice:

  • Data minimisation. Only the information a task needs reaches a model.
  • Encryption of information in transit and at rest.
  • Access controls that limit what each AI tool, and each person, can reach.
  • Audit logging of AI actions, so that they can be reviewed later.
  • Answers grounded in approved sources. Where an assistant answers from your organisation's information, we ground it in documents you approve (retrieval-augmented generation) rather than in the model's general knowledge alone.
  • Guardrails and cost controls that limit what an AI system may do, what it may say and what it may spend.
  • Operator duties. When we run AI on information your organisation is responsible for, we act as your operator under sections 20 and 21 of POPIA: we process the information only on your instructions, and we notify you immediately if we have reasonable grounds to believe that an unauthorised person has accessed or acquired it.

Accuracy, limits and the risks we manage

A model can produce fluent text that is wrong, and a well-designed workflow assumes it sometimes will. These are the main risks, in our own work and in the systems we build, and how we manage each one.

RiskWhat can happenHow we manage it
Inaccurate outputA model produces fluent, confident text that is wrongThe model is chosen for the task, answers are grounded in approved sources, and a person reviews anything critical before it is acted on
Over-reliancePeople accept a suggestion without checking itApproval points designed into the workflow, so a person confirms before anything with consequences happens
Data exposurePersonal or confidential information reaches a model or a person it should notData minimisation, access controls, encryption and POPIA controls, with a private deployment where the data requires one
ManipulationInstructions hidden in a document or message steer an AI system (prompt injection)Guardrails, limited access for each AI tool, logging, and human approval for any action with consequences
Uncontrolled costUsage grows faster than the budgetCost controls set for each system
DriftResults change as models, data or ways of working changeLogs are reviewed, and use expands only as confidence builds

Information on this website, including anything drafted with AI assistance, is general information and not professional advice. Our website terms of use explain how to read it.

How we govern AI in the systems we build for you

When we build AI into a client's systems through our AI consulting and implementation service, governance is part of the design from the first conversation. The same six steps apply whether the work is an assistant, document processing or an automated workflow.

  1. Assess readiness

    We review your data, systems, use cases and policies, and say plainly what is feasible now and what needs preparation first.

  2. Set the boundaries

    We agree in writing what the AI may see and do, and which decisions stay with your people.

  3. Choose the model and where it runs

    A cloud model, or a private deployment that keeps data in South Africa, depending on the data and your policy.

  4. Build with controls

    Guardrails, logging, access controls and cost controls go in from the start, with an approval point wherever an action has consequences.

  5. Pilot, measure, expand

    We start with a low-risk use, measure accuracy with your team and extend only what works.

  6. Report and hand over

    You receive the documentation your governance needs, covering data flows, model choices, approval points and logs, in a form your board, auditors and Information Officer can use. You own your data, credentials and documentation, and we hand them over in order when an engagement ends.

Our AI governance service helps organisations set the same policies, controls and reporting for AI they buy or build elsewhere.

For boards, public bodies and schools

Organisations that answer to a board, an auditor, a council or a governing body can ask us for the following during procurement, and managed clients can ask at their monthly service review:

  • This disclosure, and a written description of each AI feature in our proposal: what it does, the data it uses, the model, where it runs and where people approve its work.
  • Our POPIA operator terms for any service that processes personal information on your behalf.
  • Evidence for your King IV Principle 12 reporting on technology and information governance; our guide to King IV and technology governance explains what boards must show.
  • Audit-ready documentation that fits the PFMA and MFMA procurement and reporting disciplines of departments, public entities and municipalities.
  • For schools, confirmation that learner and parent data stays under the school's control as the responsible party and is never used to train models without the school's explicit consent. See cybersecurity, POPIA and online safety for schools.
  • Our company documents: we are a Level 1 B-BBEE contributor (sworn affidavit available on request), and our company registration documents, bank confirmation letter, CSD supplier number and SARS tax compliance status are available on request. We complete standard bidding documents (SBD forms) when a bid requires them, and our company facts page lists the details.

National rules on AI are still taking shape; our guide to South Africa's national AI policy tracks where the policy stands, and our POPIA compliance checklist covers the law that already applies.

Your choices and rights

  • Ask for a person. Whichever route you started on, you can reach our team on 010 447 2445 or at info@arketechnologies.com, Monday to Friday 08:00–17:00 and Saturday 09:00–13:00 (SAST).
  • Ask how AI was used. Our Information Officer will tell you whether AI was involved in handling your information or your enquiry, and how.
  • Object or withdraw consent. Under POPIA you may object, on reasonable grounds, to processing based on a legitimate interest (section 11(3)), and you may withdraw consent you have given at any time.
  • See and correct your information through the procedures in our privacy notice and PAIA manual.
  • Complain to the Information Regulator if you are not satisfied with our answer; our privacy notice gives the Regulator's contact details.

Changes to this disclosure

We update this disclosure when we adopt a new use of AI or change the providers we rely on, and when the law changes; the date at the top shows the current version. If a change materially affects how we use personal information, we update our privacy notice at the same time.

Related services

Related guides

Related work

Frequently asked questions

Does Arke Technologies use AI? Show answer

Yes, in defined places and under one rule: AI recommends and people decide. We use AI-assisted tools for Server Hub sourcing and quoting, a customer assistant whose replies staff can approve before they are sent, and AI-assisted advisory in AtlasOS. We also build AI into clients' systems. Every AI action is logged and reviewable, and this page lists each use and its control.

Does Arke Technologies train AI models on our data? Show answer

No. We never train AI models on client data without the client's explicit consent. When a system we build uses your information, we send a model only the data its task needs, encrypt it in transit and at rest, limit who and what can reach it, and log what the AI does. When we act as your operator under POPIA, we process that information only on your instructions.

Which AI models does Arke Technologies use? Show answer

We build with models from Anthropic (Claude) and Google (Gemini, through Vertex AI), and with open-weight models that can run on your own hardware. We choose the model for each job on accuracy, cost and where the data may be processed. Naming a provider does not imply a partnership. Our Information Officer can tell you which provider handles a particular service.

Will I know when I am dealing with AI rather than a person? Show answer

Yes. Transparency is one of our five AI principles, and we do not present an AI assistant as a person. Whatever route you start on, you can reach a person on 010 447 2445 or at info@arketechnologies.com during office hours. Our Information Officer will also tell you, on request, how AI was used in handling your enquiry.

Can AI make a decision about me without a person involved? Show answer

No. We do not make decisions about you based solely on automated processing that have legal consequences for you or affect you to a substantial degree, which section 71 of POPIA restricts. AI may draft, summarise or recommend, but a person makes the decision, and riskier automated changes to client systems wait for a recorded human approval.

Can our AI processing stay inside South Africa? Show answer

Yes, where your data requires it. We design private AI deployments that keep sensitive data in South Africa, running open-weight models on servers and AI workstations supplied through Server Hub. Where a cloud model is the better fit, any processing outside the country takes place only as section 72 of POPIA allows, and your proposal states where each AI feature runs.

How do we raise a question or a concern about Arke's use of AI? Show answer

Write to our Information Officer at info@arketechnologies.com, or call 010 447 2445 during office hours. We explain how AI was used, correct information that is wrong and record any objection you make. If you are not satisfied with our answer, you may complain to the Information Regulator; our privacy notice gives its contact details.

Questions about how we use AI?

Our Information Officer answers questions about AI and personal information at info@arketechnologies.com, Monday to Friday 08:00–17:00 and Saturday 09:00–13:00 (SAST). If you are planning AI for your own organisation, our team can start with an AI readiness session.

Ask about our AI useBook an AI readiness session