A finance director receives a convincing email from the chief executive requesting a time-sensitive payment. The wording reflects the organisation’s usual tone, the sender appears familiar, and a follow-up voicemail seems to confirm the request. What once looked like a crude impersonation attempt is now a plausible business interaction assembled at speed.
AI phishing trends are changing the economics of social engineering. Attackers can produce credible messages, translate them fluently, tailor them to a recipient’s role and imitate voices without the time, language skills or research effort previously required. For UK organisations, the practical question is not whether artificial intelligence has made phishing possible. It is whether existing controls can still identify and contain a more convincing, higher-volume form of it.
Why AI phishing trends matter to security leaders
Phishing remains an initial access route for ransomware, business email compromise, account takeover and data theft. AI does not replace established techniques such as credential-harvesting pages, malicious attachments or invoice fraud. It improves the presentation around them.
This matters because many older detection assumptions were based on obvious warning signs: poor grammar, generic greetings, awkward phrasing and inconsistent formatting. Those signals have not disappeared, but they are no longer dependable. Generative AI can create polished, context-specific emails in seconds, including messages written in credible British English and adapted for a particular sector.
The risk also extends beyond email. Attackers increasingly combine email with collaboration platforms, text messages, QR codes and phone calls. A message that begins in Microsoft Teams may direct a user to reset a password on a convincing sign-in page, while a voice clone is used to add urgency. Each channel supports the others, reducing the chance that an employee pauses to challenge the request.
For CISOs and IT directors, this is a control-design issue rather than simply an awareness issue. Security awareness training still has a clear role, but employees should not be the last or only line of defence against a well-resourced, AI-assisted campaign.
The AI phishing trends changing the attack surface
Better business email compromise
Business email compromise has always relied on credibility. AI helps attackers create it at scale by analysing publicly available material, previous breached correspondence or information shared across social channels. A criminal can use this material to mimic an executive’s writing style, reference a supplier relationship or construct a believable payment scenario.
The most damaging cases are not necessarily highly technical. They often exploit a legitimate business process: changing bank details, approving an invoice, releasing payroll information or sharing sensitive documents. The relevant control is therefore not just email filtering. It includes finance approval workflows, verification procedures and appropriate segregation of duties.
Multilingual and sector-specific lures
International organisations and UK firms with distributed workforces have long been targeted with translated phishing emails. AI makes translation faster and more natural, reducing the linguistic clues that recipients may have recognised in the past.
It also enables tighter sector targeting. A healthcare organisation may see messages related to patient records or procurement. A law firm may receive a document-sharing request that appears to come from a client. Manufacturing teams may be targeted with purchase orders or logistics updates. The technical delivery mechanism can be common across campaigns, while the pretext is tailored to the business function.
Deepfake voice and video verification
Voice cloning changes the risk around telephone verification. A call from a senior colleague has traditionally been viewed as a sensible way to validate an unusual payment request. It remains useful, but only if the recipient independently uses a known contact route rather than responding to a number supplied in the suspicious message.
Deepfake video is less common than AI-generated email, but it deserves proportionate planning in organisations where senior leaders, treasury teams or helpdesks handle high-value actions. Helpdesk staff are particularly exposed where attackers use synthetic audio to persuade them to reset credentials or alter multi-factor authentication details.
Faster campaign variation and evasion
Attackers can generate many variations of the same lure, changing wording, subject lines and formatting to avoid simple reputation or keyword-based controls. They may also use benign cloud services, compromised legitimate accounts or QR codes that shift users from a managed laptop to a personal mobile device.
This does not make detection impossible. It does mean that security teams should favour controls that assess identity, sender behaviour, message context, URL destination and post-delivery activity, rather than relying on a single static indicator.
Where traditional controls can fall short
A secure email gateway remains a core control, particularly for blocking known malicious domains, attachments and impersonation attempts. However, its effectiveness depends on configuration, identity integration and what happens after a message reaches the inbox. A polished message from a compromised supplier account may pass technical checks that would stop a spoofed domain.
Similarly, multi-factor authentication reduces the value of stolen passwords, but not all MFA methods offer the same resistance to adversary-in-the-middle phishing. Organisations protecting high-value systems should assess phishing-resistant approaches, conditional access policies, device posture and privileged access controls. The right mix depends on user population, application estate, operational constraints and tolerance for friction.
Security awareness programmes need updating as well. Training that focuses mainly on spelling mistakes and suspicious attachments teaches an outdated mental model. Staff should be taught to recognise unusual intent: an unexpected payment request, a change in process, pressure to bypass approvals or a request for credentials. Simulated campaigns should reflect the threats employees actually face, without turning training into a compliance exercise disconnected from day-to-day work.
A practical response for UK organisations
The strongest response combines preventative controls, identity assurance, user reporting and a rehearsed incident process. It should be designed around the routes an attacker is most likely to use, not around a generic product checklist.
Start by reviewing the organisation’s highest-risk workflows. Finance, payroll, HR, IT support, procurement and executive administration are common targets because they can authorise money movement, disclose sensitive information or change access. Document how each team validates exceptions, who can approve them and what happens when a request arrives through an unusual channel.
Then assess whether existing technology provides sufficient coverage across email, identity, endpoints and SaaS applications. Email security should identify impersonation, suspicious links, malicious attachments and anomalous sender behaviour. Identity controls should limit the impact of credential theft and flag risky sign-ins. Endpoint detection and response can help contain malicious payloads or post-compromise activity if a phishing attempt succeeds.
When comparing platforms, procurement teams should avoid treating AI capabilities as a feature in isolation. Vendors increasingly describe their products as AI-powered, yet the useful buying questions are more specific:
- Can the platform detect supplier-account compromise and executive impersonation, not only spoofed domains?
- Does it integrate with the organisation’s identity provider, email environment and security operations tooling?
- What evidence does it provide to analysts when it quarantines or permits a message?
- Can policies be tuned for different risk groups without creating an unmanageable administrative burden?
- How are false positives handled for business-critical correspondence and external document sharing?
These questions expose meaningful trade-offs. A highly aggressive policy may reduce malicious email but delay legitimate supplier communications. A platform with extensive detection capability may require specialist administration that a lean IT team cannot sustain. Managed detection, email remediation and response workflows may be more appropriate where in-house security operations capacity is limited.
Detection and response must connect
A user report of a suspicious message should trigger more than a mailbox-level investigation. Security teams need to determine whether the same sender targeted other users, whether anyone clicked, whether credentials were entered and whether those credentials were used elsewhere. This is where integration between email security, identity logs, endpoint telemetry and XDR can materially reduce investigation time.
Incident plans should also account for business email compromise without malware. If an attacker gains access to a legitimate mailbox, there may be no malicious attachment to isolate. The response may involve revoking sessions, resetting credentials, reviewing forwarding rules, identifying affected conversations and notifying suppliers or customers. Finance teams may need a rapid path to halt or recall payments.
Centralised risk registers are useful here. They allow security, compliance and operational leaders to record phishing-related risks, assign owners, track remediation and demonstrate that controls are reviewed against an evolving threat. For regulated organisations, that evidence can be as valuable as the technical capability itself.
Making a defensible technology decision
There is no single product category that solves AI-assisted phishing. The appropriate architecture depends on existing Microsoft or Google controls, identity maturity, endpoint coverage, security operations capability and the organisation’s most valuable workflows. Some businesses need to strengthen email protection first; others will gain more by improving phishing-resistant authentication, SaaS visibility or incident response coordination.
An independent assessment can help separate useful capability from broad marketing claims, particularly where several tools overlap. ITR Cyber can support organisations in comparing specialist vendors against their operational requirements, commercial constraints and deployment model, rather than forcing a decision around one manufacturer’s preferred architecture.
The most useful next step is to test a real business process before an attacker does. Ask whether a convincing message, call or QR code could bypass the controls around a high-value action - then give the relevant team the technology, authority and procedure to stop it.

