Skip to content

Your cart

Your cart is empty.

Add a service in the shop and it waits here until you check out.

Browse the shop

After you pay

  1. The team reads your details and emails you, usually within one working day.
  2. You follow the work and talk to the team in your client portal.
Subtotal

One-time service fees in USD.

Checkout View cart
AI threats

AI-written phishing: why the old warning signs no longer work

AI now writes convincing, personal phishing in any language, at scale. What the FBI, NCSC, Europol and ENISA warn, and how to train staff and harden email.

Published: 5 min read9 sources

For years, the advice was simple: look for bad spelling, odd grammar and a generic “Dear customer”. That advice is now out of date.

AI writes clean, natural, personal messages in any language, in seconds, for as many people as the attacker wants. The FBI says these tools “can correct for human errors that might otherwise serve as warning signs of fraud.” The typo that used to give a scam away is gone.

What the official warnings say

This is not one vendor’s marketing. Every major authority says the same thing.

  • FBI. In December 2024, the FBI warned that criminals use AI-written text “to appear believable to a reader” in phishing and fraud, and that generative AI “reduces the time and effort criminals must expend to deceive their targets.”
  • FBI, 2025 figures. The FBI’s report on internet crime in 2025 counts AI-linked crime for the first time: about $893 million in losses, including business email compromise that used AI-generated emails and voice cloning. The FBI says the true figures could be higher.
  • UK NCSC. The National Cyber Security Centre judges that AI “will almost certainly increase the volume and heighten the impact of cyber attacks”, and that it gives attackers a capability uplift in social engineering, making phishing more effective and harder to spot.
  • Europol. Europol’s 2025 internet crime assessment reports a rise in generative AI and large language models being used to make social engineering more effective, with scam messages tailored to victims’ culture and personal details with “alarming precision”.
  • ENISA. The EU’s cybersecurity agency found phishing was the leading way in, at about 60 percent of the intrusions it observed between July 2024 and June 2025. It cites reports that by early 2025, AI-supported phishing made up more than 80 percent of observed social engineering activity worldwide.
  • Google. Google’s threat intelligence team found that almost every notable criminal AI tool advertised on underground forums offers support for phishing campaigns, and that state-backed groups use AI to write phishing lures that impersonate people from real organisations.

What has changed

No more mistakes. Messages read like a real colleague, supplier or bank wrote them. Spelling and grammar are no longer a test.

Personal at scale. AI can read your website, your LinkedIn posts and your company news, and write a message that mentions a real project, a real supplier and the right job title. That used to take careful research by hand. Now a tool does it.

Any language. The FBI notes criminals use AI translation to remove the spelling and grammar errors that once gave foreign gangs away. A gang that cannot speak Romanian or Greek can still write clean Romanian or Greek.

Beyond email. The FBI also warns of AI-written fake websites, AI chatbots that push victims to click malicious links, and cloned voices.

Why training alone is not enough

The NCSC is clear: “No training package, including phishing simulations, can teach users to spot every phishing attempt.” With AI, that is even more true.

So the goal changes. Staff do not need to spot every fake. They need a few habits that work even when the message is perfect, and your email system needs to block what it can before anyone sees it.

Train staff on habits, not typos

  1. Check the request, not the writing. Any message that asks you to pay, sign in, open a file, share data or change details is verified through a second channel you already know.
  2. Never use the link in the message. Go to the site or app yourself, from a bookmark or by typing the address.
  3. Urgency is the signal. “Today”, “confidential”, “before the deadline”, “don’t tell anyone”. These are pressure tactics, however well written.
  4. Report fast, without blame. Make reporting one click. The NCSC stresses that a no-blame culture gets people to come forward quickly, including when they have already clicked.
  5. Refresh often. Run short sessions every few months with real current examples. The tricks change too fast for a yearly course.

Harden your email

Most of these come straight from NCSC guidance, and they work no matter how good the writing is.

  1. Stop attackers sending as you. Set up SPF, DKIM and DMARC on your domain, and move DMARC towards enforcement. This blocks the most convincing fake of all: an email from your own address.
  2. Filter at the server. Turn on your mail provider’s protection against phishing, malicious links and attachments, for every user by default.
  3. MFA on every mailbox. A stolen password alone should not open an account. For admins, use security keys or passkeys.
  4. Flag outside mail. Add a clear external-sender label, so a “message from the boss” that comes from outside stands out.
  5. Limit what you publish. The less detail about staff, roles and suppliers on your public pages, the less material for a tailored lure.
  6. Have a plan for the click. Know who resets passwords, signs people out of all sessions and checks for new mail rules, and how fast.

Where EWC fits

AI phishing methods change fast, and last year’s checklist misses them. Our security team follows new AI phishing and impersonation methods and builds them into how we set up and test your email.

Our Business email setup puts the foundation in place: email on your own domain with SPF, DKIM, an initial DMARC policy and MFA, checked and handed over to you. To see what an outside attacker could do to your website, our Advanced black-box penetration test attacks it the way a real attacker would, and reviews your email-authentication and DNS settings.

Sources

What this article is based on.

9 sources
  1. FBI Internet Crime Complaint Center (IC3)Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraudic3.gov
  2. iTnewsAI-assisted fraud makes big debut in FBI's cybercrime statsitnews.com.au
  3. NCSCThe near-term impact of AI on the cyber threatncsc.gov.uk
  4. NCSCImpact of AI on cyber threat from now to 2027ncsc.gov.uk
  5. European Commission, Migration and Home AffairsEuropol published report on how cybercriminals trade and exploit datahome-affairs.ec.europa.eu
  6. ENISAEU consistently targeted by diverse yet convergent threat groupsenisa.europa.eu
  7. ENISAENISA Threat Landscape 2025enisa.europa.eu
  8. Google Threat Intelligence GroupAI Threat Tracker: Advances in Threat Actor Usage of AI Toolscloud.google.com
  9. NCSCPhishing attacks: defending your organisationncsc.gov.uk

Where we can help · Email

Business email setup

Get email on your business domain, with account protection and delivery checks.

What you get

  • Email on your own domainOne new mailbox and up to two aliases.
  • Domain & account protectionSPF, DKIM, initial DMARC & MFA setup.
  • Your working mailbox & handoverSend/receive checks + admin handover.