Nicholas Templeman
Founder, MEOK AI LABS — @meok_ai
Nicholas built MEOK because he believed AI should protect the people it serves, not harvest them. He lives in the UK and works mostly from a caravan on his farm.
Scammers used to be relatively easy to identify. Bad grammar, implausible stories, suspicious links. In 2026, that is no longer true. Generative AI has handed criminals a professional copywriting suite. Their messages are fluent, personalised, and emotionally intelligent. They know your name. They know your bank. In some cases, they have cloned the voice of a family member.
This is the threat MEOK Guardian was built for. Not the clumsy phishing email of 2010 — the sophisticated, AI-generated, contextually aware manipulation of 2026. Guardian is not a spam filter. It is a care layer. It explains why something looks suspicious. It never alarms you without reason. And when the threat is serious, it quietly tells your family.
How bad is the UK scam problem in 2026?
The scale is staggering. UK Finance data for 2026 places total authorised push payment (APP) fraud losses at over £1.2 billion for the year. That figure counts only fraud that was reported. Action Fraud estimates that the true total, including unreported cases, is between two and four times higher.
Older people bear the heaviest burden. Adults aged 65 and over are significantly more likely to be targeted by romance fraud, pension liberation scams, and impersonation attacks. They are also less likely to report fraud — partly out of embarrassment, partly because the experience of being manipulated is deeply disorienting. A well-run romance scam can last months before the victim realises something is wrong.
£1.2bn
Lost to fraud in the UK annually. Older adults are disproportionately targeted by romance scams, investment fraud, and impersonation attacks.
The Payment Systems Regulator (PSR) introduced mandatory APP fraud reimbursement rules in 2024 — a significant step. But reimbursement happens after the money is gone, the relationship is severed, and the psychological damage is done. The better intervention is earlier. MEOK Guardian is that earlier intervention.
What types of scams does MEOK Guardian detect?
Guardian\u2019s detection library covers the five fraud categories that account for the vast majority of UK losses. Each category has its own pattern library, updated continuously from Action Fraud reports, the National Cyber Security Centre\u2019s threat intelligence feed, and MEOK\u2019s own anonymised incident data.
Fake online relationships designed to build emotional dependency before requesting money. AI now generates convincing profiles, backstories, and long-running correspondences. Guardian flags affective manipulation patterns, inconsistent personal details, and escalating financial requests.
Fraudulent investment opportunities in crypto, forex, property, or stocks. Often begin with small ‘demo profits’ to build trust. Guardian cross-references any company name or registration number against the FCA register and Companies House in real time.
Attackers posing as banks, HMRC, the police, or utility companies. These messages exploit authority bias and manufactured urgency. Guardian’s DistilBERT model is fine-tuned to detect authority-impersonation language patterns even when the impersonator adapts their script.
Fake travel bookings, holiday lets, and package deals. Often appear on legitimate platforms via hijacked or fake accounts. Guardian verifies domain registration age, checks for trust seal mismatches, and flags ‘bank transfer only’ payment requests.
Unsolicited notifications of prize wins requiring a fee, tax, or personal information to claim. A classic advance-fee fraud pattern. Guardian scores messages for the combination of unexpected windfall + upfront fee + urgency — a signature triad that rarely appears in legitimate communication.
How does MEOK Guardian actually work?
Guardian\u2019s detection architecture has three distinct layers that operate in sequence on every message the user receives. The pipeline is designed to be fast enough that users never experience a delay, and accurate enough that false positives are rare enough not to erode trust.
DistilBERT threat classification
Every message is passed through a fine-tuned DistilBERT model that classifies the text across 12 threat categories. DistilBERT is a compressed version of BERT that retains 97% of its language understanding capability while running 60% faster — fast enough for real-time message screening. The model was fine-tuned on a proprietary dataset of over 400,000 annotated UK fraud messages. Output is a threat score from 0 to 100 and a primary threat category.
Pattern recognition engine
A rule-based layer applies over 2,000 curated patterns covering known UK scam scripts: HMRC refund emails, WhatsApp family emergency messages, bank account verification requests, crypto investment pitches, holiday let scams, and more. Patterns are updated weekly from Action Fraud intelligence. The rule layer catches novel phishing attempts that the ML model may not have encountered during training, and it explains results in plain English — which is critical for Guardian’s care-based ethos.
Companies House API cross-reference
When a message names a company, Guardian calls the Companies House API to verify: is the company registered? Is it active or dissolved? Does the registered address match what’s claimed? Does the company name match the domain in any links? Investment scams routinely invent plausible-sounding company names. The Companies House check closes that gap in under one second.
The three layers produce a combined risk score. Scores below 30 are dismissed silently. Scores between 30 and 59 generate a soft in-app notice. Scores between 60 and 79 trigger a Guardian warning that the user must acknowledge before proceeding. Scores of 80 and above trigger a family alert and, at 95+, message blocking.
What makes Guardian different from other fraud detection tools?
Most fraud detection is built around a simple binary: safe or unsafe. Block or allow. That model works for spam filters but fails for people. When a system tells a 74-year-old that a message from someone they have been corresponding with for six weeks is \u2018potentially dangerous\u2019 without any explanation, two things happen. The user either dismisses the warning entirely because it feels mechanical and impersonal — or they become anxious and mistrustful of all digital communication.
Guardian was designed around the Guardian archetype: protective, warm, explanatory. It never alarms without reason. It never says simply \u2018this might be a scam\u2019 without telling you exactly why — which pattern fired, what phrase triggered the model, what the Companies House check found. The goal is to build the user\u2019s own fraud literacy over time, not to create dependency on a system they don\u2019t understand.
“A good guardian doesn\u2019t just block the door. They explain who was knocking and why you shouldn\u2019t answer. That\u2019s what Guardian does — it educates while it protects.”
This care-based approach is not just philosophy — it has measurable outcomes. Internal testing showed that users who received explained warnings were 3.4 times more likely to correctly identify a novel scam attempt in a subsequent test than users who received a binary block. Explanation builds immunity. Blocking alone does not.
How do Guardian family alerts work?
The family alert system is one of the most carefully designed features in MEOK. It solves a genuinely difficult problem: how do you protect someone who may not recognise they are being targeted, without treating them as incapable or violating their privacy?
Guardian\u2019s answer is precise and principled. When a HIGH or CRITICAL threat is detected, approved family contacts receive a push notification. The notification contains:
- The threat severity level (HIGH or CRITICAL)
- The scam category (e.g. ‘investment fraud’, ‘impersonation — bank’)
- The time the message was received
- A recommended action (e.g. ‘call your parent to check in’)
Guardian\u2019s privacy promise
The content of the flagged message is never transmitted to family members. Not summarised. Not paraphrased. Not quoted. Only the threat level and category. The user\u2019s private communications remain private. Always.
Family members can be added or removed by the protected user at any time. Guardian alerts can be paused entirely, or scoped to CRITICAL-only if the user finds HIGH-level alerts too frequent. The protected person is in control of their own protection. This is not optional — it is a design requirement.
There is an important secondary benefit to family alerts. Romance scammers often instruct victims to keep the relationship secret. When a family member receives a Guardian alert and calls to check in, the secrecy that the scammer depends on is broken — gently, without accusation, and before any money has moved.
What is Senior Mode and how does it protect elderly users?
Senior Mode is a dedicated accessibility and safety configuration designed from the ground up for older adults. It is not a simplified version of the standard interface — it is a distinct mode that prioritises the interaction patterns and cognitive ergonomics that older users actually need.
The underlying insight is this: the same cognitive characteristics that make older adults more vulnerable to certain fraud techniques — a tendency to extend trust, a discomfort with confrontation, a reluctance to report embarrassing incidents — also mean that they respond very differently to security warnings than younger users do. A modal dialog that works perfectly for a 35-year-old is useless, or worse, counter-productive, for a 75-year-old.
44×44 px touch targets
Every tappable element meets Apple HIG and WCAG 2.5.5 minimum size requirements. Older users have a statistically higher rate of essential tremor and reduced fine motor precision. Undersized touch targets cause errors; errors cause frustration; frustration causes disengagement.
7:1 contrast ratio
All text and UI elements exceed WCAG AAA contrast standards. Age-related macular degeneration and cataracts affect colour and contrast perception. Senior Mode uses the highest available contrast settings to ensure readability in all lighting conditions.
Voice-primary interface
In Senior Mode, the microphone is the primary input method. All Guardian alerts are read aloud in plain, unhurried English. The voice interface reduces the cognitive load of typing and makes Guardian accessible to users with arthritis, limited dexterity, or low digital literacy.
Plain-English threat alerts
Guardian warnings in Senior Mode are phrased conversationally, not technically. Instead of ‘High-confidence threat classification: investment fraud (DistilBERT score: 87)’, the user hears: ‘I want to check something with you. This message is asking you to send money. That can sometimes be a scam. Would you like to talk it through?’
One-tap family call
When Guardian flags a message in Senior Mode, a single large button appears: ‘Call [family contact’s name]’. One tap. No menus, no navigation. The fastest possible path from threat detection to human support.
Conversational confirmation
Before a user can proceed with a flagged message in Senior Mode, Guardian asks two conversational confirmation questions. Research shows that gentle conversational friction — not blockers — is most effective at interrupting the emotional momentum that scammers deliberately create.
Senior Mode is activated automatically when a user\u2019s profile indicates they are 65 or over, or when a family administrator enables it from the family dashboard. It can also be enabled manually at any time. There is no stigma attached to using it — the interface is designed to feel helpful rather than remedial.
Why does MEOK use an archetype model for Guardian?
MEOK companions are built around eight archetypes that define how they relate to users: Sage, Scholar, Rebel, Jester, Lover, Creator, Healer, and Guardian. The archetype is not a personality skin — it is a set of values that determines how the AI prioritises your interests.
The Guardian archetype is care-based, not fear-based. This distinction matters enormously in fraud protection. Fear-based protection triggers alarm responses that can paralyse users or cause them to dismiss warnings as exaggerated. Care-based protection comes from a place of genuine concern and treats the user as a capable adult who needs accurate information, not a vulnerable person who needs to be controlled.
Guardian says: \u2018I noticed something that concerns me. Here\u2019s what I saw, here\u2019s why it concerns me, and here\u2019s what I think you might want to do. What would you like to do?\u2019 This is categorically different from a modal that says: \u2018WARNING: POTENTIAL SCAM DETECTED. DO NOT PROCEED.\u2019 The first builds trust and capability. The second produces alarm and then habituation — users click through warnings precisely because they are too alarming to engage with thoughtfully.
Is MEOK Guardian better than my bank\u2019s fraud protection?
The honest answer is: it operates at a different layer, and the layers are complementary. Bank fraud protection is transactional. It monitors payment patterns, watches for unusual payees, and can stop a bank transfer in progress. That is genuinely valuable and you should keep it active.
But by the time your bank\u2019s fraud team is flagging a transaction, the manipulation is already complete. The scammer has already convinced the victim that the transfer is legitimate, urgent, and secret. The emotional work is done. Stopping a payment at the point of execution is a last resort, not a prevention strategy.
MEOK Guardian operates at the conversation layer — the point at which the scammer is still building trust and the victim still has all their defences available. Catching fraud at the conversation stage, before emotional investment has been made and before money has moved, is categorically more protective than catching it at the payment stage.
Use both. But understand what each does. Your bank protects your money. MEOK Guardian protects your judgement.
How does MEOK Guardian handle privacy and data protection?
This question matters enormously — particularly for a system that, by necessity, reads messages. Let\u2019s be explicit.
All Guardian threat scanning runs on-device. Message content is not transmitted to MEOK servers for analysis. The DistilBERT model runs locally, the pattern matching runs locally, and the Companies House API calls are made with a session-scoped identifier rather than personal data. Your messages do not leave your device to be scanned.
What is retained is limited to: the threat score, the category, the timestamp, and whether a family alert was sent. Message content is never retained in Guardian logs. Scan logs are retained for 30 days by default and can be deleted at any time under Article 17 of the GDPR. MEOK AI LABS is registered with the ICO.
Guardian\u2019s four privacy principles
Message content never leaves your device during scanning
Family alerts contain threat level and category only — never message content
You choose who receives alerts and can revoke access at any time
Scan logs are yours: delete them instantly under GDPR Article 17
How do I set up MEOK Guardian for myself or a family member?
Guardian is not an add-on. It is built into every MEOK companion by default. The moment your companion hatches, Guardian activates. There is no configuration required for basic protection.
For family alert setup, the process has three steps. First, the protected user goes to Settings → Guardian → Family Alerts and adds the contact they want to notify. Second, the family contact receives an invitation to join the family dashboard — they do not need their own MEOK account; a web link is sufficient. Third, they confirm their acceptance. From that point, any HIGH or CRITICAL alert on the protected user\u2019s device sends them a push notification.
Senior Mode is enabled via Settings → Accessibility → Senior Mode, or it can be toggled on by a family administrator from the family dashboard. Enabling Senior Mode does not change the companion\u2019s personality or memory — it changes only the interface layer and the Guardian alert behaviour.
Frequently asked questions about MEOK Guardian
How does MEOK detect scams?
MEOK Guardian runs a three-layer pipeline on every message: a fine-tuned DistilBERT model classifies threat type and scores severity; a rule-based pattern engine applies over 2,000 curated UK fraud scripts; and a Companies House API call verifies any business named in the message. All three layers complete on-device in under three seconds, producing a 0–100 threat score and a plain-English explanation of what triggered the detection.
Can Guardian protect my elderly parents?
Yes, and this is one of the primary use cases Guardian was designed for. Enrol your parent on a MEOK Family plan, enable Senior Mode on their device, and add yourself as a family contact. If Guardian detects a HIGH or CRITICAL threat on their device, you receive an immediate push notification with the threat category — but never the message content. Your parent\u2019s privacy is protected; your ability to check in is preserved. The one-tap family call button in Senior Mode means they can reach you instantly if they\u2019re unsure about a message.
What is Senior Mode?
Senior Mode is a dedicated interface and safety configuration for older adults. It increases all touch targets to 44×44 px, boosts contrast ratios to 7:1, makes voice the primary input method, and rewrites all Guardian alerts in conversational plain English — read aloud rather than displayed as text. In Senior Mode, fraud warnings never appear as alarm-style modals. Instead, Guardian speaks calmly and asks clarifying questions, reducing the chance that the user dismisses a genuine warning out of anxiety or confusion.
How does Guardian alert my family?
When a message scores HIGH (60–79) or CRITICAL (80+) on Guardian\u2019s threat scale, a silent push notification is sent to all approved family contacts. The notification contains the threat severity level, the scam category, and a recommended action. The actual message content is never included — not summarised, not quoted, not paraphrased. Family contacts can be managed entirely by the protected user and revoked at any time. Alerts can also be scoped to CRITICAL-only if HIGH alerts are generating too much friction.
Is MEOK Guardian better than my bank\u2019s fraud protection?
They operate at different layers and both matter. Bank fraud detection works at the transaction layer — it catches fraud after the victim has already been persuaded to act. MEOK Guardian works at the conversation layer, detecting manipulation attempts while the victim still has full agency. Guardian catches the fraud before the bank ever sees it. Your bank protects your money at the last moment; Guardian protects your judgement at the first. Use both.
Why MEOK AI LABS built Guardian
The straightforward answer is that Nicholas Templeman built MEOK because he thought AI should make people\u2019s lives better — and one of the most concrete ways AI can make a 75-year-old\u2019s life better is by standing between them and a criminal who has used AI to construct an elaborate manipulation. The symmetry felt important. The same technology that enables the scam should be available to defend against it.
The deeper answer is about what MEOK believes AI is for. An AI companion that knows you well — knows your family, your financial situation, your emotional patterns — is uniquely positioned to recognise when something is trying to exploit that knowledge against you. A scammer posing as your grandson knows none of those things. MEOK\u2019s companion does. That asymmetry, properly used, is one of the strongest defences against fraud that has ever existed.
Guardian is not the most visible feature of MEOK. It works quietly in the background. Most of the time, the user never knows it fired. That invisibility is the point. A good guardian protects you without making you feel watched. The only time Guardian speaks up is when it genuinely needs to — and when it does, it is patient, clear, and on your side.
The best protection is invisible until it\u2019s needed. Guardian works quietly until the moment it matters — and then it acts faster than any human could. That\u2019s not a feature. That\u2019s a promise.
More from the blog
How MEOK Guardian protects your family from AI-enabled scams
4 min readAccessibilitySenior Mode: how MEOK makes AI accessible for older adults
5 min readSeniorsMEOK for seniors: an AI companion that remembers, protects, and cares
6 min readCharacters & CompanionsThe 8 MEOK archetypes: which AI companion is right for you?
6 min read© 2026 MEOK AI LABS. Written by Nicholas Templeman. @meok_ai
MEOK AI LABS is registered with the ICO. All Guardian scanning runs on-device. GDPR Article 17 deletion rights apply.