AI Hyper-Personalised Phishing Scams

Traditional phishing attacks are generic: the same email goes to millions of recipients, and most people recognise the format immediately. AI-powered spear-phishing removes this genericness entirely. Automated tools can now scrape your name, employer, recent social media posts, professional history, colleague names, and geographic details from public sources, then generate a phishing message that mentions specific real details — making it almost indistinguishable from a genuine communication.

Where conventional phishing relies on shock or urgency with no personalisation, AI-generated phishing feels like a natural continuation of your actual professional or personal life. The email might reference a project your LinkedIn shows you are working on, mention a colleague by name, or relate to a news story relevant to your employer. This specificity dramatically increases click-through rates and credential submission.

As AI language models become more capable and more accessible, the cost and skill required to run these attacks approaches zero. What previously required significant human research effort can now be automated at scale.

The attacker first runs an automated profile-scraping process against the target. Public sources include LinkedIn, corporate websites, social media, press releases, news articles, conference speaker lists, and professional directories. From these, a detailed picture of the target emerges: their role, their projects, their colleagues, their clients, and their recent activities.

An AI language model then generates a phishing email that incorporates these details naturally. The email might purport to be from a named colleague asking for a document review, from an industry publication with a personalised newsletter that requires a login, or from a vendor related to a project the target is known to be working on.

The link in the email leads to a credential-harvesting page, a malware download, or a fake login portal. Because the message references real context, the recipient's guard is lower than it would be for a generic phishing attempt.

At the enterprise level, AI also enables automated pretexting — building relationships through multiple emails or messages before introducing the harmful element, all without human effort on the attacker's side.

Generic phishing is effective against inattentive recipients. AI-personalised phishing is effective against careful recipients because it removes the usual signals that trigger scepticism. When an email knows your name, your project, your colleague's name, and your employer, the automatic assumption is that it is genuine — because fabricating that level of detail previously required effort that mass-phishing campaigns could not justify.

The psychological gap between a suspicious-looking generic phishing attempt and a message that reads like natural workplace communication is enormous. AI bridges that gap at scale, without the effort that would previously have limited such attacks to high-value targets only.

Hi [name], I saw your recent post about [project]. I have a relevant document to share — view it here: [fake link].

Following up from [conference name] last month — I wanted to send over the slides we discussed. Review at [fake portal].

Hi [name], [colleague name] mentioned you would be the right person for this. Can you review [document] by end of day? [fake link].

Your account on [professional platform] requires verification following recent changes. Confirm your details: [fake link].

For any message that prompts you to click a link, log in, or share a file — regardless of how genuine it appears — verify the request through a separate channel. Call the apparent sender, send a new message to their confirmed address, or speak to them in person before acting.

Hover over links before clicking to see the destination URL. Legitimate corporate documents and portals will be hosted on domains you recognise and expect.

Be aware that the level of personalisation in a message is no longer evidence of legitimacy. Treat unexpected requests for credentials or document access with the same scrutiny whether they appear generic or specific.