Hyper-Personalized Phishing: When AI Knows Your Org Chart Better Than Your New Hire
Some phishing emails now sound so much like your boss that you would reply without a second thought. Others sound like a vendor you pay every month. On 1 June, one automated email campaign reached more than 67,000 people across 42,000 organizations in under three hours. Microsoft tracked it. The emails posed as sales heads first and then as the CEO.
AI phishing attacks begin with research, and the research is easy. LinkedIn pages, press releases and job ads show who reports to whom, which vendors you use and who signs off on payments. AI then writes an email that sounds like someone the reader trusts. It can copy a voice on a phone call too. This is called hyper-personalized phishing, and it is happening now.
AI Phishing Attacks Study Your Company Before They Write
Before an attacker writes a single email, they read about the company. None of it needs hacking. Four kinds of public information do most of the work.
Your Team Pages Show Who Reports to Whom
Company websites and LinkedIn pages list names, titles and managers. They also show who joined recently, and a new joiner is an easy target. They do not yet know who normally asks for what, so a request from the CFO feels normal.
Your News and Vendor Announcements Show Who You Pay
Press releases and news posts name new vendors, big contracts and new offices. That gives the attacker a name that looks normal in your finance team’s inbox.
Your Job Ads Name the Systems Your Staff Use
A job ad for a SAP finance analyst tells the attacker which system finance uses and which login page to copy.
Your Webinars and Talks Can Give Away a Voice
Recorded webinars, talks and interviews put leaders’ voices online. An attacker can copy a voice from a short clip.
Put these facts together and you can see who approves payments, who can grant access and who trusts whom.
Turning Public Information Into a Targeted Phishing Attack
A language model handles the writing. It copies how the person being imitated writes, and it can make a different version for each target. Aimed at one named person, this is AI-powered spear phishing. Used on many employees, it becomes AI social engineering at a scale manual research could never reach. AI phishing attacks like these follow four steps.
Step 1: The attacker picks the person who can say yes
They look for a person who can pay, approve or give access. That is usually someone in finance, IT, HR or an executive’s office.
Step 2: They build a believable pretext
The reason matches the person’s daily work. It looks like any other request.
Step 3: They copy someone the target trusts
This might be a spoofed address, a lookalike domain or a hijacked real account. It can also be a cloned voice or a fake video message.
Step 4: The target takes the action and the attacker settles in
The employee enters credentials on a fake page, makes a payment, sends a file or approves a request. With a stolen login, the attacker can read the mailbox, set up forwarding rules and quietly watch for the next payment.
Why “Spot the Typo” Training Is No Longer Enough
For years, staff were given the same checklist. Look for spelling mistakes. Watch for “Dear Customer.” Be careful with strangers. It was good advice, and it still works on mass phishing emails. It does not help much against AI phishing attacks, which are written for one person.
Here is how each tip holds up today.
- Spelling mistakes. The email is clean and matches the sender’s usual style.
- Generic greetings. The email uses your name, your job title and a project you are working on.
- Unknown senders. The email comes from a taken-over account or a near-identical domain.
- Strange links. There may be no link at all. The email only asks you to reply or to pay.
So the problem is not grammar. These emails carry real details about your business, and that makes them look genuine. Staff are being asked to catch something that was built to look normal.
Common AI Phishing Attacks That Reach Finance, IT and Leadership
These attacks go after money, authority and access. Each one lands on a different desk.
A Supplier Asks for New Bank Details and Finance Pays the Wrong Account
The email arrives in a real supplier thread, or from an address that looks almost the same. It says the supplier has changed banks and asks finance to update the details. It looks routine. Once the money goes to the wrong account, it is very hard to get back.
The CEO Calls With an Urgent Payment and the Voice Is Fake
A voice note or a short video call seems to come from the CEO. It says a deal is closing today and a payment must go out now. The rush leaves no time to check. A cloned voice or face like this is called deepfake phishing, and it does not stay in email. The Verizon’s 2026 DBIR found that 41% of social engineering breaches used a route other than email, such as phone calls and social media.
An IT Admin Signs In to a Fake Page and the Attacker Gets In
An IT admin gets a renewal notice for a tool the team really uses. The page looks right, so the admin signs in. Now the attacker has the login or an open session, and an admin account reaches far more than one mailbox.
In all three cases, the damage starts with one person doing a normal part of their job. AI phishing attacks like these can also grow into bigger fraud, which is where business email compromise comes in.
How a Business Email Compromise Attack Bypasses Traditional Security Controls
A business email compromise (BEC) attack often has no harmful file or link. It is only a request: pay this invoice, change this account, send me that list. Security tools that look for bad files or links have little to flag.
The FBI’s Internet Crime Complaint Center recorded about $3.05 billion in BEC losses in 2025 from 24,768 complaints. That counts only what victims reported.
AI now helps with the hardest part, which is knowing who to fake. In September 2026, Microsoft’s Digital Crimes Unit shut down EvilTokens. Its chatbot read hijacked inboxes, found who trusts whom and who approves payments, then suggested who to impersonate. So people are being asked to catch what the tools cannot.
Phishing Awareness Training Works Best With a Rule Behind It
Training still matters. A trained employee slows down before acting and reports an email that feels wrong. The first people to see a phishing email are almost always the staff it was sent to, so their habits count. Verizon’s 2026 DBIR found a human element in 62% of breaches, up from 60% a year earlier.
But training has a limit. No one can reliably spot a perfect fake, even a careful finance approver at quarter end. What helps is a rule that works even when the email looks real.
- Check bank changes by phone. Confirm any change to a supplier’s bank details on a number already on file, never one from the email.
- Treat urgent requests the same way. A payment marked “urgent” gets the same check, even if it seems to come from the CEO.
- Add a second approver. A second person signs off on any new bank account before money moves, and the check is built into the finance system so nobody can skip it.
Microsoft gave the same advice after the EvilTokens case: check payment changes through a trusted second channel. A rule only works if everyone follows it, and on a busy day someone will not. That is why a company also needs systems that catch what a person misses.
Layered Security Catches What a Single Control Misses
A rule can be skipped and a person can be fooled, so a company needs several checks.
Layer | What it catches | What can still get past it |
Email security | Fake sender addresses, lookalike domains, first-time senders, risky links | A taken-over real account sending a clean-looking email |
Sign-in security | Logins from unusual places, new sign-in methods, stolen sessions | A real employee approving a convincing request |
Payment checks | Bank detail changes and fake executive requests | Anyone who skips the rule |
Monitoring (SOC) | Email, sign-in and device alerts in one picture | Not much, but it needs people watching around the clock |
Monitoring ties the rest together. One new forwarding rule looks harmless. So does one login from a new country. A next-gen SIEM shows them as one story, and a managed SOC has people to read it at 2 a.m. Most teams cannot build every layer at once, so the order they start in matters.
Phishing Prevention Checklist for IT Teams: Sign-In, Email, Testing and Monitoring
The first controls protect the accounts attackers want most. The last one catches whatever gets through.
1. Phishing-resistant sign-in. Put security keys or passkeys (FIDO2) on finance, IT admin and leadership accounts first. Turn off SMS codes on those accounts.
2. Email authentication. Set up SPF and DKIM, then move DMARC from monitor to reject once the reports look clean. Tag outside emails so staff can see them.
3. Realistic phishing tests. Send finance a fake supplier bank change and IT a fake licence notice. Track who reports it and how fast, not only who clicked.
4. 24×7 monitoring. Watch for new forwarding rules, sign-ins from new countries and new devices on admin accounts. These attacks do not wait for office hours.
The same checklist works for phishing prevention for SMBs. An internal IT team can usually handle the first three. Monitoring needs people on shift all night, and that is where many companies look for outside help.
Attackers Will Keep Researching, So Make Every Request Easy to Verify
The 1 June campaign reached tens of thousands of people in under three hours. The next one will not wait for a company to feel ready. The attacker needs only a few public pages and a believable request.
A company does not have to out-guess the attacker. It has to make sure no single email, call or video can move money or open an account on its own. A good first question: what happens here if a fake request looks perfect?
One rushed approval should never be enough to drain a bank account. Progression helps make sure it is not. Our next-gen SIEM and managed SOC watch email, sign-in and device activity day and night. Endpoint, network and cloud security shut the next doors, and data protection limits the damage. Start with a review of where your email, sign-in and monitoring stand today.
FAQs
- Can AI phishing emails get past email security filters?
Yes, some can. Most filters look for bad links, bad files or known bad senders. An email that only asks you to pay an invoice has none of these, so there is little to catch. A few will still slip through, so staff checks and monitoring still matter. - Does MFA protect against AI phishing?
Only some kinds do. A text code can be tricked out of a person, or the session can be stolen after login. Security keys and passkeys only work on the real website, so a fake page gets nothing. Use them on finance, IT admin and leadership accounts first. - How do you tell if an email was written by AI?
Most of the time you cannot, and you do not need to. Look at what the email asks for. If it wants money, new bank details, a password or a rushed decision, check it by calling a number you already have. - Why are small businesses also at risk from AI phishing?
The research costs the attacker almost nothing. A LinkedIn page, a press release or a job ad is enough to build a believable request. Smaller firms often have fewer payment checks too, so one fake email is more likely to work. - What should an employee do right after approving a fake payment?
Tell IT security straight away, then call the bank and ask them to stop or recall the payment. Even a few hours’ delay makes recovery harder. IT should end the person’s sessions, reset the password and check for new forwarding rules.
