What Is Local AI? A Guide for Directors Handling Confidential Documents

Most directors now reach for AI to help with a dense board pack. Fewer have asked where those confidential papers actually go the moment they do.

Local AI is artificial intelligence that runs entirely on your own device, rather than on an outside company's servers. It reads, summarises and answers questions about your documents — much as ChatGPT does — but nothing travels over the internet and nothing is stored by anyone else. For market-sensitive, privileged or commercially confidential board material, that single architectural difference is the whole point.

Key takeaways

• Local AI runs on your own computer and sends nothing to an outside provider — the documents never leave your device, so there is no external copy to leak, subpoena, or train on.
• 51% of UK professionals have entered work-related correspondence into public AI tools, according to PagerDuty's 2026 Shadow AI survey, fielded by the independent pollster Wakefield Research.
• The UK's National Cyber Security Centre warns that queries typed into public tools like ChatGPT are visible to the provider and "almost certainly" used to develop the model.
• A 2026 UK tribunal found that putting confidential documents into a public AI tool such as ChatGPT can waive legal privilege — the first English ruling to address the point directly.
• Local AI removes that exposure by design: no external server, no retained copy, and no internet connection required to do the work.

What is local AI, in plain terms?

Local AI is AI that processes your documents on the same device you are sitting in front of, instead of sending them to a data centre run by someone else. The technical name for the underlying idea is edge inference — defined by Wikipedia as "running machine learning or deep learning models on local devices... instead of centralized cloud computing infrastructure," with "improved privacy by reducing the amount of data sent to remote servers" listed as a core advantage.¹

You will see three terms used for roughly the same thing. On-device or on-premise AI describes where the work happens — your machine, not an external server. Private AI describes the result — your material stays confidential. A local AI model is simply the AI itself, packaged to run on ordinary hardware rather than in a data centre. A tool that is genuinely local is private, because the documents never leave the device to begin with.

The distinction only matters for some documents. For drafting a public speech or researching a published report, where the material travels is academic. For a board pack containing pre-market results, an M&A discussion, or privileged legal advice, it is a governance question — and that is where local AI earns its place.

How is local AI different from cloud AI like ChatGPT?

The difference is not the intelligence; it is the address the intelligence works at. With a cloud service such as ChatGPT, Gemini or Copilot, your documents are transmitted to the provider's external servers, analysed there, and the result is sent back. With local AI, the model runs on your own device and the documents stay put.

That routing has consequences most users never see. The UK's National Cyber Security Centre put it plainly in its guidance on large language models: "The query will be visible to the organisation providing the LLM (so in the case of ChatGPT, to OpenAI)... Those queries are stored and will almost certainly be used for developing the LLM service or model at some point," according to NCSC authors Dave Chismon and Paul J, 2023.² For a director, "the query" may be an entire confidential board paper.

We have written a full, factual comparison in Local AI vs Cloud AI. In brief:

Why does it matter for confidential board work?

Because the way professionals actually use AI has run well ahead of the controls around it. In PagerDuty's 2026 Shadow AI survey, 31% of office professionals said they had entered financial information or confidential company documents into public AI tools, according to fieldwork by the independent firm Wakefield Research.³ Among UK respondents specifically, 51% had put work-related correspondence into such tools — higher than the four-country average of 43%.³

Directors are not exempt from this pattern; if anything they are at the sharp end of it. OnBoard's 2026 Board Effectiveness Report found that 92% of board directors had used AI for board work in the previous six months, yet 60% of them sat on boards with no formal AI policy, according to OnBoard, 2026.⁴ Separately, BlackFog's research, conducted by the independent agency Sapio Research in late 2025, found that 49% of employees had used AI tools their employer had not sanctioned.⁵ (OnBoard and BlackFog both sell governance and security products, so treat their figures as informed industry research rather than neutral academic study — but the direction is corroborated by the independently-fielded PagerDuty numbers above.)

The consequences are no longer hypothetical. In 2026 a UK tribunal held, in the Munir case, that uploading confidential documents into an open-source AI tool such as ChatGPT places that information in the public domain and can waive legal privilege — reported as the first English tribunal ruling to address the point directly, and distinguishing closed, enterprise tools operating inside a secure network as lower-risk.⁶ As Olivia Morton, a Professional Support Lawyer at the UK firm Birketts, framed it: "AI tools should be thought of as open meeting rooms rather than locked filing cabinets," Birketts, 2026.⁷

For a director, that is the crux. Privilege waived by careless handling can undermine a litigation position; a board paper exposed through a provider's breach or a discoverable chat log engages your personal responsibility for confidentiality. Local AI closes the exposure at the architectural level: material that never leaves your device cannot be discovered on someone else's servers.

What can local AI actually do for a director?

The same substantive work you would want from any capable assistant — applied to the documents you cannot risk sending away. A local tool can read a 300-page board pack, summarise the material risks, cross-reference the finance report against the strategy paper, and surface the questions worth asking on Tuesday — all on your own machine.

This is where meetinginsight.ai fits. It runs a local AI model on your own device, so you can add board papers and interrogate them in plain language, with the analysis and the documents both staying on the machine in front of you. There is no account into which papers are sent and no external server holding a copy. For directors juggling two, three or four boards, the same tool works across all of them without mixing contexts or requiring each company's separate sign-off. If your interest is the preparation method itself, our guide to reading a board pack covers the workflow in depth.

How can you tell if a tool is genuinely local?

The word "local" is used loosely in marketing, so it pays to know how to check. Four questions settle it:

• Does it analyse documents with the internet switched off? Genuine local AI does. If it stops working when you disconnect, material is being sent somewhere.
• Must you create an account and sign in before it will read a document? A required login often means the document is routed through an external service to be processed.
• Does the provider's privacy policy mention retaining, reviewing or training on your content? With true local processing there is nothing for them to retain, so there is nothing to disclose.
• Is there a single, unambiguous answer to "does anything in my documents leave this device?" If the answer is qualified — "encrypted before sending", "processed securely" — the documents still leave.

A tool that passes all four keeps your board papers where they belong. One that does not is, to some degree, a service wearing local clothing.

Is local AI as capable as cloud AI?

Honestly, the largest external models still lead on the most open-ended, general tasks — and it would be misleading to claim otherwise. But two things are true at once. First, the capability gap has closed quickly, which we cover in our companion explainer, What Is a Local LLM?. Second, for the specific job of reading and reasoning over your own documents, a modern local model is more than equal to the task; you rarely need a frontier model to summarise a board pack accurately.

One nuance worth stating, in the spirit of facts over reassurance: "local" is only as private as the tool's design makes it. Running a model on your device is the right foundation, but a well-built, dedicated offline application — one that genuinely keeps documents and analysis on the machine and sends nothing out — is what delivers the confidentiality, not the label alone. The right question to ask any vendor is simple and unambiguous: does anything my documents contain ever leave this device? With true local AI, the answer is no.

In summary

Local AI is AI that runs on your own device, so confidential documents are analysed without ever being sent to an outside company. For a director handling market-sensitive, privileged or commercially confidential board material, that is not a technical preference — it is the difference between a private working session and placing papers on infrastructure you do not control. The intelligence is comparable; the exposure is not.

If you want to understand the technology underneath, read What Is a Local LLM?; if you want the head-to-head, read Local AI vs Cloud AI. And if you would rather simply use it, meetinginsight.ai analyses your board papers entirely on your own device — nothing sent, nothing stored elsewhere. Try a free 30-day trial at meetinginsight.ai/download.

Notes

---

meetinginsight.ai processes your board papers entirely on your device. Nothing sent. Nothing stored elsewhere. Download a free 30-day trial at meetinginsight.ai/download.

Footnotes

1. Wikipedia, "Edge inference." https://en.wikipedia.org/wiki/Edge_inference ↩

2. Dave Chismon and Paul J, National Cyber Security Centre (UK), "ChatGPT and large language models: what's the risk?", 14 March 2023. https://www.ncsc.gov.uk/blog-post/chatgpt-and-large-language-models-whats-the-risk ↩

3. PagerDuty, "Shadow AI Workplace Survey 2026," fielded by Wakefield Research (1,250 office professionals across Australia, Japan, the UK and the US, April 2026). https://www.pagerduty.com/blog/ai/shadow-ai-workplace-survey-2026/ ↩ ↩²

4. OnBoard, "2026 Board Effectiveness Report," May 2026. (OnBoard is a board-management software provider; figures are vendor research.) https://www.onboardmeetings.com/blog/boards-are-using-ai-the-risk-is-how-onboard-launches-ai-suite-for-responsible-governance/ ↩

5. BlackFog, "Shadow AI Threat Grows Inside Enterprises," fieldwork by Sapio Research, November 2025 (published January 2026; 2,000 employees, half UK, half US). https://www.blackfog.com/blackfog-research-shadow-ai-threat-grows/ ↩

6. UK and R (on the application of Munir) v Secretary of State for the Home Department (AI hallucinations; supervision; Hamid) [2026] UKUT 00081 (IAC) — a joined Hamid-jurisdiction hearing, cited in legal commentary variously as Munir v SSHD and UK v SSHD. As reported by Norton Rose Fulbright, "Court guidance that use of open-source AI waives confidentiality and legal professional privilege," April 2026, and Birketts (below). https://www.nortonrosefulbright.com/en/inside-disputes/blog/202604-court-guidance-that-use-of-open-source-ai-waives-confidentiality-and-legal-professional ↩

7. Olivia Morton, Birketts, "Using AI and the risk of waiving legal privilege," 1 April 2026. https://www.birketts.co.uk/legal-update/using-ai-and-the-risk-of-waiving-legal-privilege/ ↩