How to Train Employees to Spot Vishing: Train the Callback, Not the Ear

Mandiant’s M-Trends 2026 found that “highly interactive voice phishing saw a significant surge to 11%, becoming the second-most commonly observed vector” in its 2025 investigations, while email phishing “dropped to just 6% of intrusions in 2025.” The 2026 Verizon DBIR found that phone-centric phishing simulations, which count text messages, voice calls and callback emails together, had a median click rate “closer to 2%,” against 1.4% for email. That rests on only 35 campaigns, and the authors wrote that they “struggled to find companies doing simulations of voice- and text message-based campaigns.”

So, what’s the best way to train people for it? Our answer: stop trying to teach employees to hear a fake. Teach a verification habit, train the help desk on right procedures, rehearse both on real phones, and measure reports as carefully as failures.

Why listening for a fake caller doesn’t work

Most vishing awareness content is a list of things to listen for: a strange accent, background noise, an unknown number. On the phone those cues are weak. The New York Department of Financial Services’ February 2026 vishing advisory describes attackers posing as IT help desk staff who “often use spoofed caller IDs when calling personnel on their personal and work phones.” Voices can be cloned too. In a study by Mai and colleagues, 529 listeners “only correctly spotted the deepfakes 73% of the time,”. And they knew they were listening for fakes.

Retool’s breach in August 2023 shows the deeper problem. In Retool’s own account, the caller “claimed to be one of the members of the IT team, and deepfaked our employee’s actual voice,” and was “familiar with the floor plan of the office, coworkers, and internal processes.” The employee “grew more and more suspicious” and still handed over one more MFA code. Recognition happened, and it didn’t stop anything.

That’s how vishing is built to work. A review of 86 real attacks by Jones, Armstrong, Tornblad and Siami Namin found that “authority (A), social proof (S) and distraction (D) were the most widely used persuasion principles in vishing attacks.” Those work on people in real time because mid-conversation there’s no pause to think. What cannot be disguised is the request itself and that’s what training should focus on.

What the evidence says about vishing training

We’d love to point at a study proving vishing training has a lasting effect in a workplace. There isn’t one. The closest is a field experiment by Bullée and colleagues, published in 2016, in which fake callers phoned 48 employees. Of those who saw an awareness campaign a week before the call, 9.1% followed the caller’s instructions, against 46.15% of those who saw nothing. Two weeks after the campaign, the effect was gone: 54.6% complied. A newer pilot by Hashmi, George and colleagues, presented at CHI 2023, tested students instead. They trained 100 university students, half on how scammers use authority and half with general awareness, then called them unannounced eight days later, with a caller claiming to be a university employee dealing with an IT problem. Of the 50 who picked up, “26 subjects (52%) complied with the scammer for at least the first step.” The persuasion-trained group did a little better, 47.6% against 55.2%, and the difference wasn’t statistically significant.

Here’s how the same paper sums up the earlier work: “the evidence of effective training in raising awareness for vishing is yet to be shown.” It also cites a study by Scheibe and colleagues in which “subjects were more likely to detect vishing calls for example contexts that were included in the training, as opposed to those for a new context.”

Email research shows similar bleak pictyre. The UC San Diego Health trial of more than 19,500 employees (Ho et al., IEEE S&P 2025) found an embedded-training effect that was statistically significant “but of only 2%.”

Taken together, that’s a case against relying on awareness. Knowledge of tricks fades within two weeks. What’s worth training is a simple rule tied to the request, practised often enough that it survives even the most convincing caller.

Make verification the skill you train

On this, the authorities agree almost word for word. The FBI’s December 2024 warning tells people to “verify the identity of the person calling you by hanging up the phone.” Google’s Threat Intelligence Group, in its hardening guidance on the UNC6040 vishing campaigns, asks organisations to teach “hanging up and calling the official account manager using a phone number on file,” routing requests through the official support portal, and asking for “a valid ticket number that can be confirmed in the support console.” NYDFS says: “instead of relying on Caller ID, implement procedures for personnel to confirm the identity of individuals” asking for access.

So the training content is short. Employees learn the rule and the requests that trigger it. The triggers come straight from recent campaigns:

• a one-time code or an MFA approval, as in the Scattered Spider calls described in the CISA and FBI advisory
• a password or MFA reset
• a remote support session, as in the fake IT calls Microsoft documented in May 2024
• approval of an app or connector, as in the UNC6040 Salesforce calls
• a change to payment details or an urgent transfer

Teach the setups as well as the requests. In the campaigns Microsoft describes, attackers first flood an inbox with subscription spam and then call offering to fix it. An unexpected IT call about an unexpected problem is much more suspicious.

When one of those requests comes up, the employee hangs up and calls back on a number from the directory, or asks for a ticket they can check. Nobody has to judge the voice. The hard part is permission. Say clearly that hanging up on “the CFO” or “IT” is always acceptable, and have the real executives say it too.

Train the help desk as its own audience

Ordinary staff get called by fake IT. This type of attacks resulted in some of the most expensive cases. Clorox’s lawsuit against its help desk contractor, as reported by The Record, alleges that “at no point during any of the calls did the Agent verify that the caller was in fact Employee 1.” The 2023 attack on MGM Resorts followed the same route, if the attackers’ own widely reported (and unconfirmed) account is right: find an employee on LinkedIn, then call the help desk.

A help desk agent’s job is to help, so a habit alone won’t hold up against a caller who sounds locked out and desperate. Agents need a procedure, and their training is rehearsing it. Google’s guidance is the most detailed public version: “Do not inherently trust the caller’s stated identity,” “no single factor should be sufficient for high-risk actions,” and avoid identifiers anyone can look up, such as date of birth or a supervisor’s name. For MFA resets and privileged accounts it adds a call to “the user’s registered phone number on file” or manager approval. The UK NCSC gave similar advice after the April 2025 attacks on UK retailers, telling organisations to review how their helpdesks authenticate staff before resetting passwords.

ISO/IEC 27001:2022 Annex A 6.3 asks for training “as relevant for their job function,” which is a good argument for a separate track. And if agents are scored on how fast they can solve caller’s problem, training will fail. Make it explicit that an agent who follows it with an angry caller on the line did the job right.

Rehearse on the phone, under pressure

NYDFS’s October 2024 industry letter says social engineering training “can be effectively delivered through simulated phishing, and voice and video impersonation exercises,” and NIST SP 800-53 control AT-2(1) asks organisations to “provide practical exercises in literacy training that simulate events and incidents.”

It helps to know what each kind of exercise actually tests. A recorded call driven by keypad presses tests whether someone presses a key. A voice conversation inside a lesson rehearses reasoning about a call the employee knows is practice. A live call to a real phone, where the caller pushes back when the employee hesitates, tests whether they verify under pressure. A test call to the help desk tests the reset procedure. We’d start with disclosed practice and then move to unannounced live calls with varied pretexts, because people mostly catch the scenarios they’ve already seen (remember the Scheibe finding).

Set a few guardrails before the first live call. Get legal sign-off on recording, since some jurisdictions need every party’s consent; California Penal Code section 632 is the usual example. Get written consent from any executive whose voice you clone. And pick pretexts with care. After GoDaddy’s fake holiday bonus test, the company said “we need to do better and be more sensitive to our employees” (Engadget).

Count reports and verification, not just disclosures

Most programs report one vishing number: how many people disclosed something. That misses the behaviour you actually want to practice. The UK NCSC’s phishing guidance asks organisations to “focus on how many people reported it,” and warns that “users who fear reprisals will not report mistakes promptly, if at all.” Google’s guidance asks for “a clear and accessible process for end-users to report suspicious communications.”

Phones have no report button, so build one: a short internal number, a chat channel or a form, named in every piece of vishing training. Then track four things per exercise: who disclosed, who tried to verify (hung up, called back, asked for a ticket), who reported and how quickly, and whether the help desk followed procedure on every test call. Benchmarks are thin. The 2025 Verizon DBIR found that users trained in the previous 30 days reported simulated phishing emails at about 21%, against a base rate of 5%. But that’s email, and neither the 2025 nor the 2026 DBIR gives a reporting rate for phone simulations. Your own trend line will tell you more.

Fake voices will keep getting better, and that’s fine. A callback to a known number works the same whether the voice is real, cloned or just a confident actor.