An accounts payable employee receives an email from a long-time vendor. It references a real invoice and a real project, and it asks to update the vendor’s banking details before the next payment run. The writing is clean and professional. There are no typos, no strange formatting, and nothing that looks wrong. The employee makes the change, and the next payment goes to an account the attacker controls.
Nobody in that scenario was careless. The email simply gave the employee nothing to catch.
That is the problem AI phishing creates for leadership. Here is the question every CFO, CEO, and owner should be able to answer: if an employee received a flawless message from you, or from a trusted vendor, asking them to move money or share credentials, what in your company would stop it that does not depend on the employee noticing something odd?
The stakes are measurable. The FBI’s 2025 Internet Crime Report, released in April 2026, found that phishing and spoofing was again the most reported cybercrime, with 191,561 complaints. Total reported cybercrime losses reached a record $20.9 billion, up 26% from 2024. Business email compromise alone accounted for more than $3 billion. For the first time, the report included a dedicated section on artificial intelligence: more than 22,000 complaints and about $893 million in losses tied to AI. The FBI has warned that generative AI reduces the time and effort criminals need to deceive their targets and corrects the human errors that used to give fraud away.
What Is AI Phishing and How Is It Different From Traditional Phishing?
CISA describes phishing as criminals trying to get people to open harmful links, emails, or attachments, delivered by email, text, social media message, or phone call. AI changes the speed, scale, quality, and personalization of that attack.
AI phishing is a phishing attack that uses artificial intelligence to write, personalize, and scale deceptive messages, calls, or videos. The goal is still to steal money, credentials, or access, but the message is fluent, tailored to a specific person and company, and free of the mistakes that used to give scams away.
Traditional phishing was often a mass message with generic wording, and the sloppiness was part of what gave it away. AI phishing attacks can be tailored to a specific person, role, and company, written in polished language, and produced in minutes. What the attacker wants has not changed. How convincing the attack is has changed dramatically.
How Attackers Use AI to Make Phishing Attacks More Convincing
The FBI has publicly described how criminals use generative AI for fraud, and the methods fall into a few clear patterns.
First, attackers build personalized messages from public information. Company websites, press releases, and professional profiles give them real names, projects, and vendor relationships, so the email reads like it came from someone who knows your business.
Second, AI produces flawless writing and translation. That removes the grammar and spelling mistakes employees were taught to watch for, and it lets attackers write convincingly in any language.
Third, attackers create supporting material that makes an impersonation believable: fake websites, fictitious social media profiles, and AI-generated images and identification documents.
Fourth, voice cloning lets an attacker impersonate an executive or known contact in a phone call or voicemail, using only a short sample of that person’s voice.
Fifth, deepfake video lets an attacker appear as someone else on a live call, or pass as a job candidate in a remote interview.
Finally, AI gives attackers volume and speed. They can produce and adjust large numbers of tailored messages with very little effort, then refine whatever works.
The most striking example came from the engineering firm Arup. In January 2024, a finance employee in Arup’s Hong Kong office joined a video call with what appeared to be the company’s UK-based CFO and several colleagues. Every person on the call was an AI-generated deepfake built from publicly available footage. The employee had initial doubts about the original email request, but set them aside because the people on the call looked and sounded like colleagues he knew. He made 15 transfers totaling about $25.6 million to five bank accounts. Arup confirmed the incident in May 2024.
The lesson for leadership is direct. A video call is no longer proof of identity, and any process that treats it as proof can be beaten.
Why Traditional Phishing Advice No Longer Works
Much of the standard phishing guidance still leans on spotting errors: misspelled words, awkward grammar, and unusual formatting. Those signs can still appear, and CISA still lists them, but they are no longer a reliable test. Training built around those cues can leave employees confident while the attacks they actually receive contain none of them.
Other defenses have weak points too. Basic email filters tuned to known bad patterns can miss a message that is new and well written. And a policy that depends on individual judgment puts the full burden on whoever happens to read the message, often while busy and under time pressure.
The shift leadership needs to make is from relying on instinct to relying on process.
What AI Phishing Attacks Look Like Inside a Real Business
Most AI phishing attacks against businesses follow a few familiar patterns, and each one is tied to a business outcome.
A vendor payment change is the most common. A convincing email asks accounts payable to update banking details, and the next legitimate payment goes to the attacker.
Executive impersonation is the Arup pattern. A message, call, voicemail, or video call that appears to come from the CEO or CFO asks for an urgent wire and asks the employee to keep it quiet.
Credential harvesting uses a realistic login page for a common business platform to capture a password. From there, the attacker moves into email and shared files and looks for the next opportunity.
Payroll and HR requests come from someone posing as an employee who wants to change direct deposit details before the next pay cycle.
Each of these works because it uses the company’s normal workflows. Business email compromise losses typically move through wire transfers and ACH payments, the same processes the finance team uses every day.
The Business Cost of an AI Phishing Attack
The direct loss from a fraudulent payment is only the start. If a stolen password leads to a wider breach, the business faces recovery costs and downtime. If client data is exposed, it may face notification and legal obligations. And client relationships can suffer long after the money is gone.
The FBI’s 2025 figures show the scale. Business email compromise losses exceeded $3 billion, and businesses reported more than $30 million in BEC losses with a confirmed AI connection. Reported phishing and spoofing losses grew to about $215 million, roughly triple the prior year. The FBI notes that these numbers are underreported.
These losses also connect to pressures leadership already feels. Enterprise clients send security questionnaires that ask about awareness training and email controls. Cyber insurance applications ask about multi-factor authentication and payment verification procedures. Compliance audits look for documented training and incident response. A phishing incident that exposes a missing control quickly becomes a conversation about contracts and coverage, not just IT.
How to Spot AI Phishing: Judge the Request, Not the Writing
Employees cannot be expected to detect every AI phishing message by reading it closely. The more useful habit is to focus on what a message is asking for rather than how it looks. Several signs still hold up.
Be cautious of pressure or urgency, especially around money, credentials, or gift cards. Watch for any request to change how something is normally done, such as new banking details or a new payment method. Treat a request to bypass the usual approval process, or to keep something secret, as a serious warning. Check whether the sender address or link domain matches the real organization, even when the display name looks right. And notice a sudden move to a different channel, such as an email followed by a text, a phone call, or an unexpected video meeting.
The FBI also recommends listening closely to tone and word choice on calls and watching for small imperfections in images and video. Those clues are getting weaker as the technology improves, so they should never be the only line of defense. For teams that want to go further on the technical side, our guide on how to spot a phishing email using headers walks through what to check.
Why This Can’t Wait
The urgency here is tied to deadlines most businesses already have. The next cyber insurance renewal will likely ask about multi-factor authentication and verification procedures for payment changes. Many policies limit coverage for social engineering or funds transfer fraud, and some require documented callback procedures, so it is worth asking your broker what your policy actually says before renewal. The next enterprise client security questionnaire will ask about training, email security, and incident response. The next audit will look for documented training and response procedures. And for defense contractors, CMMC includes security awareness training requirements that cover phishing and social engineering.
Each of those is a moment when someone outside the company will ask whether these controls exist, and “we trained our people” will not be enough. Every payment run without a verification step is another chance for one convincing message to succeed. Reported fraud losses rose 26% in a single year, and attackers only need to be lucky once.
How to Stop AI Phishing: Controls That Work Even When the Message Looks Perfect
The strongest defense against AI phishing is a set of business processes that hold up no matter how convincing the message is. Each one is a leadership decision, not a technical setting.
Verify Through a Separate Channel
Any request to change payment details or move money should be confirmed through a known phone number or another separate channel, never by replying to the request or using the contact details it contains. After Arup, that rule needs to include requests made on video calls.
Require Dual Approval
Set a dollar threshold above which payments need a second approver. That way, one deceived employee cannot complete a transfer alone.
Turn On Multi-Factor Authentication
Multi-factor authentication on email and critical systems means a stolen password alone is not enough to get in. Stronger, phishing-resistant options are available and worth considering for high-risk accounts.
Strengthen Email Security
Filtering and domain authentication reduce spoofed messages reaching inboxes. SPF, DKIM, and DMARC are standards that help confirm an email really came from the domain it claims, and they make it harder for attackers to impersonate your company.
Limit Public Exposure
The FBI advises limiting the online availability of executives’ images and voices. Reducing what attackers can learn about your company and leadership from public sources makes personalized attacks harder to build.
Document Your Incident Response
A written phishing response plan should say who to call and how quickly. If a payment goes out, contact the bank immediately and file a complaint with the FBI’s Internet Crime Complaint Center. In 2025, the FBI froze about $679 million in fraudulent transfers when it was alerted in time. Our ransomware preparedness and incident response planning service helps businesses build and test these plans before they are needed.
Training and Reporting: Building a Team That Catches What Technology Misses
Technology will never catch everything, so employees remain an important layer of defense. But training has to match the threats they actually face. That means phishing simulations that mimic realistic AI-generated attacks, refreshed regularly, and training that teaches the verification process rather than only listing warning signs. Building everyday security habits across the whole team, not just IT, is covered in our article on cybersecurity for non-tech teams.
Reporting matters just as much. CISA’s guidance on recognizing and reporting phishing stresses making it easy and safe to report, even after someone has already clicked. If employees fear blame for reporting a mistake, attackers get more time. A fast, no-blame reporting process is one of the cheapest and most effective controls available. As we cover in One Click Away: Cyber Resilience for Small Businesses, a single click reported in minutes is a near miss, while the same click hidden for days can become a serious incident.
What Leadership Should Be Asking
- If an employee received a perfectly written request to change vendor banking details or send a wire, what would stop it?
- Do we require verification through a separate channel before payment details or credentials change, and is that written down?
- Would our finance team still verify if the request came on a video call that looked like me?
- Is multi-factor authentication turned on for email and every system that holds sensitive data?
- When was our last phishing training, and does it reflect AI-generated attacks or only the old warning signs?
- Can employees report a suspicious message in one click, and do they feel safe doing so?
- Do we have a documented plan for the first hour after someone clicks or a payment goes out?
- Could we answer a client or insurer questionnaire about phishing controls today with documentation?
How eSilo Helps
eSilo helps leadership turn phishing defense from a hope that employees will notice into controls and training the business can document and defend. Every eSilo advisor brings 15+ years in large enterprise or federal government environments, where payment controls, access controls, and incident response are designed and tested, not assumed. An email filter is one layer, but it does not fix a payment approval process.
eSilo helps businesses:
- Assess exposure to phishing and impersonation, including payment, credential, and access workflows
- Build verification and approval procedures for high-risk requests
- Deliver employee training and realistic phishing simulations that reflect current attack methods
- Review multi-factor authentication and email security coverage
- Create a phishing incident response plan with clear ownership
- Prepare documentation that supports client questionnaires, insurance applications, and audits
eSilo has protected business data since 2002, with an average client relationship of more than 10 years, Best of Florida recognition from 2022 through 2025, and WOSB, MBE, and WBE certifications. Its cybersecurity awareness training prepares teams for the attacks they will actually see.
Frequently Asked Questions
What should we do if someone on our team already sent a payment to a fraudulent account?
Move fast. Contact your bank immediately to request a recall, file a complaint at ic3.gov, preserve the emails and logs, and notify your cyber insurance carrier as your policy requires. In 2025, the FBI froze about $679 million in fraudulent transfers when victims reported quickly. This is exactly why a documented response plan matters: the first hour should not be spent figuring out who to call.
What is the single most effective step we can take this month?
Require verification through a separate, known channel for any payment, banking, or payroll change, and write it down. It works no matter how convincing the message is. If possible, pair it with dual approval above a set dollar threshold.
Our employees have already had phishing training. Isn’t that enough?
Training still matters, but training built around spotting typos and odd wording prepares people for the attacks of five years ago. Effective training now teaches the verification process and is backed by controls that do not depend on anyone noticing anything.
How do we verify a call or video meeting that sounds exactly like our CEO?
The Arup case showed that even a live video call with familiar faces can be fake. The answer is process, not perception: hang up and call back on a known number, use a pre-agreed verification step for urgent financial requests, and make it clear from the top that executives expect to be verified and will never penalize anyone for checking.
Does multi-factor authentication stop AI phishing?
MFA makes a stolen password far less useful and is one of the most important controls available. It does not stop an employee from being tricked into sending a payment, though, and some phishing kits can intercept weaker forms of MFA. Phishing-resistant MFA is stronger, and MFA works best alongside payment verification.
Does cyber insurance cover losses from AI phishing or fraudulent wire transfers?
Coverage varies widely. Social engineering and funds transfer fraud are often subject to sub-limits or separate endorsements, and some policies require documented verification procedures. Review your policy with your broker before renewal so you know where you stand.
How do we know if we’re actually protected, or just think we are?
The test is whether your controls work when the message is perfect, and whether you can show documentation when a client or insurer asks. If you are not sure, that is the gap a structured review is designed to find.
Will Your Process Catch the Perfect Message?
AI has made phishing harder to detect, but it has not made it unstoppable. Businesses that stop relying on employees to spot a bad email, and instead build verification, authentication, training, and reporting into how the company operates, will be far better positioned than those that wait for an incident to force the issue.
If the next phishing message your company receives is flawless, will your process still catch it? Contact eSilo to start a Risk and Resilience Diagnostic, a structured three-week engagement that tests how your payment, credential, and access workflows would hold up against a convincing AI phishing or impersonation attempt and delivers a prioritized plan. Close the gaps before your next insurance renewal, client questionnaire, or audit asks about them.




