Written by Israel Egede

AI can help small businesses and charities save time, improve admin, and work more efficiently. Retrospectively, it brings risk such as inaccurate information, data exposure, unexpected costs and over-reliance on tools that do not behave as expected.

AI is More than ChatGPT

When people say “AI” they often mean ChatGPT, but AI is a much broader field and tools like Claude, Gemini, Kimi and Perplexity all sit within it.

AI is also not new. The ideas behind it go back decades: Alan Turing’s work on machine intelligence in 1950, early conversational systems such as ELIZA in the 1960s and later research into the transformer architecture that serves as the foundation for many modern AI tools. ChatGPT’s launch in late 2022 made these systems widely visible but it did not create AI overnight. It also is not solely a US industry as we have Google DeepMind based in London and organisations such as Mistral, Z.ai and Deepseek. They have all developed models that are available through platforms such as Hugging Face.

‍

Some of the Myths and Risks of AI

Myth 1: AI Understands Everything

AI can seem understanding because it responds naturally and confidently. But it does not understand information in the same way people do.

LLMs generate responses based on patters and probabilities so are useful for drafting, summarising, brainstorming and explaining but they can also produce plausible sounding information that is wrong, often called a hallucination.

Even the people building these systems do not fully understand every reason a model produces a particular response. Research such as Anthropic’s “Golden Gate” experiment showed how difficult it is to interpret what is happening inside these models. This is why human in the loop remains important especially for anything factual, legal, financial or sensitive.

‍

Myth 2: AI Detectors are Reliable

AI detectors cannot reliably tell whether text was written by AI so they can produce false positives and negatives example in 2023, OpenAI withdrew its own AI text classifier because it was not accurate enough.

There is also an ongoing arms race between detection tools, watermarking and methods designed to remove or avoid detection. For charities and small businesses writing funding bids, this means AI detection should not be treated as fool proof. A better focus is quality: is the bid accurate, evidence-based, tailored to the funder and reflective of your organisation?

‍

Myth 3: AI is Only for Technical Experts

AI is not only for large companies with data teams as free and low-cost tools exist and are useful for cafes, volunteer coordinators, food banks, sole traders, small charities and all other organisations.

No deep technical knowledge is needed to start. The key skill is being able to explain a task clearly: what you need, who it is for, what format you want and what a good result looks like. See AI more like a new team member you need to brief very well to get better results.

‍

Myth 4: Agents Only do What They are Told

AI agents can take actions, not just answer questions. This makes them powerful hence the need for clear boundaries.

In August 2026, the UK AI Security Institute reported concerning behaviour from an AI model during cybersecurity testing. After initial approaches failed, the model went beyond its intended task and attempted to manipulate a real third-party GitHub repository. It is a reminder that AI agents need clear boundaries, safeguards and human in the loop especially where they can access the internet, systems, data or payments.

Prompts, Projects and System Instructions

A prompt is the instruction you type into an AI tool to get a result, but prompts are only one layer.

• ‍Project instructions can set the context for a specific piece of work, such as the client, goals, tone, and allowed sources.
• ‍System instructions sit behind the scenes and set wider rules for how the AI should behave.‍
• Data controls determine what happens to the information you enter.

For organisations, this matters because a well setup AI workspace can improve consistency and reduce risk. For example, project instructions might say which documents the AI can use, what it must not include, and how outputs should be formatted.

‍

Risk 1: Hallucinations and Fake Sources

One of the clearest risks is relying on AI without checking its output.

In a well-known US legal case, lawyers submitted a court briefing containing six cases invented by ChatGPT. The court fined the lawyers and their firm $5,000 after the fabricated citations were discovered.

A similar issue arose in the UK when a taxpayer appealed an HMRC penalty by citing nine earlier tribunal decisions that did not exist. The tribunal accepted that the citations were likely generated by AI.

Even if a model is right most of the time, the remaining errors can matter most particularly in legal, tax, health, safeguarding, or financial settings. Always verify sources, citations, figures, and claims before using them.

‍

Risk 2:  Chatbots Can Make Promises

AI chatbots can create real business risks when they give customers incorrect information.

Air Canada was held liable after its chatbot told a bereaved customer that they could book a full-price flight and claim a bereavement discount afterwards. The discount policy did not exist in the way the chatbot described, and the company had to compensate the customer.

In another example, a DPD chatbot was manipulated into swearing at a customer and criticising the company. These cases show why chatbots need clear instructions, limits, human oversight, and escalation routes.

‍

Risk 3:  Jailbreaking and Automated Actions

AI systems interact through text, so this lets people manipulate them through carefully worded prompts. This is known as jailbreaking.

One example involved a chatbot on an online store being persuaded to apply a 100% discount. Because the chatbot was connected to automated ordering and payment systems, the order could be processed before the business owner noticed.

If you use AI in customer-facing systems, make sure discounts, refunds, payments, and account changes require approval or have clear limits. Automated actions should never be completely unchecked.

‍

Risk 4: Data and Confidentiality

Free AI tools may use conversations to improve their models unless data controls are changed. This creates a risk if staff paste in confidential information, client data, source code, meeting recordings, or personal information.

A useful way to think about it is the postcard analogy: information sent through a system may be visible to more than just the intended recipient. Treat anything entered in an AI tool as potentially shareable unless you know exactly how that tool handles data.

Samsung reportedly restricted generative AI use after engineers pasted confidential source code and meeting recordings into ChatGPT. In another example, an employee saw an AI-generated summary of a meeting they had left, including discussion about why they had not been successful in an interview.

Before using AI, organisations should be clear about:

• What information staff can and cannot enter?
• Whether a free or paid business plan is appropriate?
• Where is data stored and how it is used?
• Who can access AI-generated meeting summaries?
• Whether personal or special category data is involved?
• How your organisation meets UK GDPR and Data Protection Act responsibilities?

UK data protection law still applies. Organisations remain accountable for the data they put into AI tools and the information they produce with them, with oversight from the Information Commissioner’s Office (ICO).

‍

Risk 5: Runaway Costs

AI can create unexpected costs if usage is not monitored. Businesses have been caught out by runaway costs when nobody kept track of how AI tools were being used.

Set spending limits, review subscriptions regularly, assign ownership for AI tools, and monitor usage especially where tools are connected to APIs, automations, or customer-facing systems.

‍

Risk 6: Over-reliance and Sycophancy

AI tools can be overly agreeable so they may praise ideas, reinforce assumptions, or make users feel more confident than they should be. This is sometimes called sycophancy.

That can be dangerous when someone uses AI for major decisions, emotional support, financial choices, or legal matters. AI should support human judgement, not replace it.

‍

How to Use AI Safely

You do not need to avoid AI. You need to use it deliberately.

• Start with low-risk tasks such as drafting, summarising, brainstorming, and organising information.
• Keep confidential, personal, and client data out of tools unless you understand the data settings.
• Use business or enterprise plans where stronger data controls are needed.
• Always check facts, sources, citations, calculations, and legal or financial advice.
• Set clear rules for staff use, including approved tools and prohibited information.
• Use project and system instructions to set boundaries and improve consistency.
• Monitor usage and costs.
• Keep a human responsible for final decisions.

‍

The Benefits are Real

Used well, AI can reduce time spent on admin, financial reporting, content drafting, research, and repetitive processes. These are often the quickest wins for small teams.

AI can help a small organisation do more without necessarily needing to hire immediately. But the goal should not be to replace people with AI. It should be to remove friction, improve consistency, and free up time for the work that needs human judgement, relationships, and care.

‍

Final Thoughts

AI is neither magic nor something to fear. It is a set of powerful tools that can create real value but only when organisations understand their limits, protect their data, check outputs, and keep people in control. The organisations that benefit most will not necessarily be the ones using the most advanced tools. They will be the ones using AI sensibly, safely, and with a clear purpose.

Want to Use AI More Confidently?

Digital Pathfinders offers free workshops, events, and one-to-one support to help SMEs, charities, and social enterprises across the North East adopt technology safely.

Explore upcoming events and resources at digitalpathfinders.uk or get in touch to find out how we can support your organisation.

Use AI with curiosity but keep your judgement, data, and customers protected.

‍