Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
AI-assisted phishing uses generative models to write flawless, personalised lures at scale, and increasingly to clone a voice or a face, removing the clues staff were trained to spot.

What Is AI-Assisted Phishing?
AI-assisted phishing is social engineering in which a generative model does the writing, the research, or the impersonation. The attacker describes a target and a pretext; the model produces a message in correct business English, in the right register, referencing the recipient’s role and recent company news. The awkward grammar and odd formatting that awareness training spent a decade teaching people to notice are gone, and they are not coming back.
Three things. The first is quality: a non-native speaker can now produce a flawless message in any language, which removes the single most reliable visual cue. The second is scale with personalisation—techniques that previously justified the effort only for a chief financial officer can now be applied to two thousand employees, each message drawing on public profiles and company announcements. The third is channel: voice and video cloning move the attack off email entirely. The NCSC’s assessment published on 24 January 2024 concluded that AI will almost certainly increase the volume and heighten the impact of cyberattacks over the following two years, with the clearest near-term uplift in social engineering and reconnaissance rather than in novel exploitation.
Business email compromise is the obvious beneficiary, because the attack was always about plausibility rather than technical exploitation. Thread hijacking becomes far more dangerous when a model can continue an existing conversation in the sender’s own style. Help desks are a second pressure point: a cloned voice calling the service desk to request an MFA reset defeats a process built on recognising a colleague. Recruitment and supplier onboarding are a third, where synthetic identities now pass a video interview. Our notes on voice cloning fraud and deepfakes cover those channels in more depth.
The content cues are gone, so lean on the controls that never depended on them.
Successful phishing that exposes personal data is a reportable breach to the ICO within 72 hours, and payment fraud should go to Action Fraud; UK banks operate mandatory reimbursement rules for authorised push payment fraud, which makes a documented verification process commercially as well as operationally valuable. The NCSC’s Suspicious Email Reporting Service remains the route for forwarding lures. At Privia Security, we run realistic phishing and vishing campaigns against agreed scenarios as part of our Social Engineering Services.
Not by writing quality, which is why filters have shifted to signals the text cannot hide: sender reputation, authentication results, domain age, infrastructure, link behaviour, and deviation from normal communication patterns for that pair of correspondents. Detection accuracy on content alone has fallen, so treat the filter as one layer and assume well-written lures will reach inboxes.
Yes, with a different syllabus. Teaching people to look for bad spelling is now counterproductive, because a polished message reads as legitimate. Train on process instead: any request to move money, change bank details, or reset credentials gets verified out of band, regardless of who appears to be asking or how urgent it sounds. Make reporting easy and non-punitive.
Remove voice as a form of proof. Verify requests through a second channel using contact details from your own directory, agree a challenge procedure for finance and help desk teams, and require dual approval for payments above a threshold. Some organisations use a shared code phrase for executive requests; the control that matters is that no single call can authorise a transfer.
Not so far in social engineering. The techniques are familiar—pretexting, invoice fraud, credential harvesting—but the cost of producing convincing, targeted content has collapsed, and real-time voice and video impersonation is new in practice if not in concept. The NCSC’s assessment points the same way: more volume and better quality rather than fundamentally different attacks.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.