Artificial intelligence can replicate voices, identities, authority and trust at unprecedented scale. ATI helps identify the signals, behaviors and threat patterns that still matter.
ATI evaluates evidence and threat context. It does not treat “AI detection” as proof.
ATI is organised around the ways artificial intelligence changes deception. Each domain defines what is being reproduced, what the attacker gains from it, and which signals remain reliable.
Domains are additive. New categories of AI-enabled deception are introduced here as they are documented, without changing how existing intelligence is presented.
A few seconds of public audio is enough to reproduce a familiar voice. ATI treats voice recognition as a weak identity signal and focuses on what the voice is asking for.
Video of a recognisable face carries authority it may not have earned. ATI evaluates the claim the media is supporting, not just the media itself.
A fabricated persona has no real-world anchor. ATI looks for the absence of verifiable history rather than the presence of polish.
Fluent writing is no longer a trust signal. ATI analyses the structure of the request, the pressure applied and the context, not the quality of the prose.
AI is used as the claim, not the mechanism. ATI checks whether the operation behind the claim can be verified at all.
Screenshots, reviews and group chats are trivially fabricated. ATI weighs independent verification above any trust signal the sender controls.
Voice, video, screenshots, writing quality and polished online identities were once reasonable evidence of authenticity. They are now reproducible.
“A familiar voice”
can be cloned.
“A video endorsement”
can be synthesized.
“A professional profile”
can be manufactured.
“A polished conversation”
can be automated.
Operating principle
When money, credentials, codes, account access or urgent action are involved, verify through an independent trusted channel.
Concrete situations, the element AI contributes, the threat it enables and what still verifies. Each leads to the NEXO surface that analyses it.
Modern AI-enabled scams can look and sound completely legitimate. ATI demonstrations show the techniques directly and explain which signals still matter.
Synthetic demonstration·Generated for education
Sample pendingAudio sample not yet published
A fictional, generated scam call will be added here with a transcript. No real person's voice will be used.
A fictional emergency call in which a cloned voice requests an urgent transfer. Nothing in this demonstration is based on a real person.
Each dossier answers four questions: what it is, how AI is used, what the attacker is trying to achieve, and what to verify. Open one for signs, weighted signals and an illustrative scenario.
A call or voice message that appears to come from a family member, colleague or employer, built around an urgent scenario: an accident, a medical emergency, bail, an overdue business payment.
Voice-cloning tools trained on short public audio, often taken from social media, can reproduce a familiar voice well enough to pass a casual listen.
Convert recognition into compliance before the target pauses to verify. Payment is usually requested through channels that are hard to reverse: transfers, crypto or gift cards.
Hang up and call the person back on a number you already have. Ask something only they would know. Treat "don't tell anyone" and "act now" as reasons to slow down.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
"Mum, I've had an accident. My phone is broken, this is my friend's number. I need £800 now for a tow truck. Please don't tell Dad yet."
Related patterns: Vishing · Romance scam
Check a phone numberSocial media advertisements in which a recognisable public figure appears to explain how a trading platform or "AI system" generated fast returns. The link leads to a registration page requesting a deposit.
Synthetic video and audio, or manipulated real footage, place a known face and voice behind a script they never delivered.
Borrow the credibility of a familiar face to move the target onto a controlled platform where fabricated gains encourage larger deposits and withdrawals are blocked behind "fees".
Check the public figure's official accounts, which typically address fake endorsements. Search the platform in your national financial regulator's register before sending anything.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
A video ad shows a well-known businessman explaining how a "new AI trading system" earned £5,000 in a week. The link opens a registration page asking for an initial deposit.
Related patterns: Investment scam · Crypto scam · Influencer impersonation
Check the platform linkCold outreach that reads like a legitimate professional message: a consulting offer, a collaboration, a recruiter, a partnership. The scenario is plausible; the request is what matters.
Writing tools produce fluent, tailored copy at scale, referencing a target's name, employer or recent activity and adapting tone across languages.
Lower suspicion long enough to introduce a payment, a credential request, a link or an off-platform move.
Verify the sender through an independent channel. Judge the request, not the prose. An unexpected request stays unexpected regardless of how well it is written.
ATI does not conclude that a message was written by AI from its style. Polished writing is not evidence on its own; the manipulation structure, the request and the context are what ATI evaluates.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
"Hi [Name], I'm reaching out on behalf of [Company]. I came across your work on [Project] and we'd love to discuss a consulting opportunity. Could you share your availability this week?"
Related patterns: Phishing · Job offer scam · Brand impersonation
Analyze the messageA convincing platform with a live-looking dashboard. Users deposit funds to "activate" the AI, see inflated returns and are encouraged to add more. Withdrawal attempts trigger invented fees, taxes or technical issues.
AI is the product claim, not necessarily the mechanism. Terms such as "algorithm", "automation" and "trading bot" manufacture confidence before any money moves.
Sustain deposits for as long as the fabricated dashboard remains credible, then disappear.
Check the company against your national financial regulator's register directly. Do not rely on reviews, screenshots, group chats or support agents on messaging apps.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
A platform promoted in an ad claims an AI trading bot returns 3–8% daily. A group chat shows screenshots of withdrawals. A $250 deposit is required to activate the bot. On withdrawal, a second payment for "taxes" is requested.
Related patterns: Investment scam · Crypto scam
Check the platform linkA near-identical account, or a "new number", contacting the real person's friends, family or colleagues with an out-of-character request.
Profile content can be copied and reassembled quickly; voice and writing style can be imitated. The result exploits social trust that was built by someone else.
Use an existing relationship to bypass the scepticism a stranger would face, then request money, favours or sensitive information.
Contact the person through a channel you already had before this message. Ask a personal question. Do nothing until identity is confirmed.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
An Instagram DM arrives from what looks like your sister's account: "Hey it's me. I'm in trouble and need £200 quickly, I'll explain later." The account has two posts and was created last week.
Related patterns: Brand impersonation · Romance scam
Check the profileA persona with no real person behind it: a generated face, a backstory, a professional history. Commonly a romantic contact, a recruiter or a support agent.
Generated imagery and scripted conversation allow one operator to run many personas simultaneously, each building trust at its own pace.
Establish a relationship or professional pretext, then convert it into a financial request, an ID document or account credentials.
Insist on a live video call before any financial or professional commitment. Verify recruiters through the company's official website. Never share credentials with unsolicited support.
One signal rarely proves anything. Combinations, and the request they surround, raise risk.
A recruiter named "Sarah Mills" from "Global Talent Partners" sends a connection request. Her profile looks professional but has no verifiable company link. She offers a high-paying remote role and asks for your ID to "start background checks".
Related patterns: Job offer scam · Romance scam
Check the profileATI does not treat a polished message, an unusual voice or a realistic image as proof of AI involvement. Conclusions are based on supported signals, context and observable threat behavior.
Where a conclusion cannot be supported, ATI says so. An honest “not established” is more useful than a confident guess.
Alerts and briefs from the NEXO Intelligence Hub that relate to AI-enabled deception.
