Pretexting: the attacker on the phone already knows your name, your manager and your problem.

01 What pretexting is

Verizon’s definition is precise. Pretexting is a tactic “in which a trusted relationship is built through concocted scenarios to trick the user into taking an action that unknowingly compromises the organization, frequently by voice communications but also seen via email or text messaging” [ Verizon DBIR 2026 ] .

The 2026 report, the 19th edition, covers more than 22,000 confirmed breaches. It finds the human element in 62% of them, up from 60% a year earlier. Social engineering is the third most common breach pattern, at 16%. And for the first time Verizon tracks pretexting as its own initial access vector, because a significant number of high-profile ransomware breaches in its data started that way [ Verizon DBIR 2026 ] .

02 Pretexting vs phishing

Phishing sends a message and waits. Pretexting talks back. Verizon draws the line on timing: phishing is “an asynchronous social action”, while pretexting “intrinsically involves a synchronous component”, with the attacker “on the other side of the phone, text message conversation or email thread” [ Verizon DBIR 2026 ] .

The two often run together. Verizon notes that callback emails, which ask you to ring a number, and bank-detail change emails are phishing followed by pretexting: “If there is someone on the other side of the proverbial line interacting with you to do something you shouldn’t, that’s Pretexting” [ Verizon DBIR 2026 ] .

The distinction matters for your training budget. Click-simulation emails teach people to spot a bad link. They do nothing for a helpdesk agent whose job is to be helpful to a stressed caller. Verizon puts it bluntly: training helpdesks “to not be helpful and supportive” when someone is manipulating them is not as simple as checking whether an email is external [ Verizon DBIR 2026 ] .

03 How a pretext is built

A good pretext takes an attacker an hour or two to build. It has four parts.

• Research. LinkedIn gives names, job titles, reporting lines and start dates. The company website gives the helpdesk number and the name of your IT provider. Breach dumps and stealer logs give old passwords and personal phone numbers. The NYDFS found the Twitter attackers “used personal information about the employees to convince them that the Hackers were legitimate” [ NYDFS 2020 ] .
• Persona. The attacker picks a role the target already expects to hear from: a colleague, the IT team, a supplier, an auditor. The best persona is someone real who is hard to reach right now, such as a manager on holiday or a new starter nobody has met.
• Urgency. A deadline shuts down checking. The board meeting starts in ten minutes. Payroll closes today. The VPN fault is spreading. Each one makes a callback feel like an obstacle.
• Authority. The persona carries enough rank or legitimacy that refusing feels risky. A director who cannot get in. An auditor with a regulator’s name. IT staff who “need” your code to fix your account.

Put together, the call sounds routine. The person who resets the password usually believes they did their job well. Attackers skip the call altogether when an old password from a breach dump still works, so a free check of which logins open with a leaked password is worth running first.

04 Five pretexts used against smaller firms

These five show up again and again against small and mid-sized firms. Each one targets a process that exists to help people quickly.

The helpdesk reset is the one that leads to ransomware. It hands over an identity, and an identity reaches email, file shares and the VPN. Okta described the pattern in August 2023: callers tried “to convince service desk personnel to reset all Multi-factor Authentication (MFA) factors”, and in its customers’ cases went after accounts with Super Administrator rights [ Okta 2023 ] .

The bank-change pretext costs the most money. The FBI’s Internet Crime Complaint Center recorded $3,046,598,558 in reported business email compromise losses in 2025, up from $2.77 billion in 2024 [ FBI IC3 2025 ] . The same pressure tactics drive the fake-CEO payment request, which deserves its own article.

05 Real pretexting examples

Each case below comes from a regulator report, a securities filing, a court claim or the victim’s own account, and we keep to what those say.

MGM Resorts, September 2023

MGM shut down systems across its US properties in September 2023. TechCrunch reported guests finding room keys, slot machines and payment systems out of action. Scattered Spider claimed the attack, and told the research group vx-underground that it found an employee on LinkedIn and called the help desk to get into that person’s account [ TechCrunch 2023 ] . That account comes from the attackers. MGM has not publicly named the entry point.

MGM’s filing gives the cost. In its 8-K of 5 October 2023, the company estimated a negative impact of about $100 million on Adjusted Property EBITDAR, plus less than $10 million in one-time expenses. September occupancy fell to 88%, against 93% a year earlier. The attackers took names, contact details and driving licence numbers, and for a limited number of customers Social Security and passport numbers [ MGM 8-K ] .

CISA and the FBI later described the same group’s method in a joint advisory: using voice calls “to convince IT help desk personnel to reset passwords and/or MFA tokens”, and posing as IT staff to get employees to share one-time codes [ CISA AA23-320A ] .

Caesars Entertainment, September 2023

In the same month, Caesars told the SEC its breach resulted from “a social engineering attack on an outsourced IT support vendor used by the Company”. The attacker took a copy of its loyalty programme database, including driving licence and Social Security numbers for a significant number of members [ Caesars 8-K ] . Note the word outsourced. Your helpdesk may be someone else’s staff.

Clorox, August 2023

In July 2025 Clorox sued Cognizant, its outsourced help desk, over its 2023 attack. According to the claim, attackers phoned the help desk on 11 August 2023 and had Okta and VPN passwords and Microsoft MFA reset, and the phone number on the employee’s MFA changed. “At no point during any of the calls did the Agent verify that the caller was in fact Employee 1,” the filing says. Clorox claims $380 million in damages. Cognizant denies it and says it did not manage Clorox’s cybersecurity [ The Record 2025 ] . These are allegations in an open case.

Twitter, July 2020

The New York regulator’s report is the clearest public record of a pretext call. On 14 July 2020 the attackers phoned several Twitter employees and “claimed to be calling from the Help Desk in Twitter’s IT department”, responding to a VPN problem. Remote working had made VPN faults common, so the story fit. Employees typed their credentials into a lookalike VPN site, the attackers replayed them live, and some employees approved the MFA prompt that followed. In roughly 24 hours, 130 accounts were compromised [ NYDFS 2020 ] .

Marks & Spencer, April 2025

M&S chair Archie Norman told MPs in July 2025 that the attack began with a “sophisticated impersonation” involving a third party. M&S estimated the hit to operating profit at £300 million [ Bloomberg 2025 ] . Days after the retail attacks began, the NCSC told organisations to review “how the helpdesk authenticates staff members credentials before resetting passwords, especially those with escalated privileges” [ NCSC 2025 ] .

06 What AI voice cloning changes

For years, “I recognise the voice” worked as a rough check. It no longer does. In August 2023 an attacker texted Retool staff about a payroll problem, got one employee to log in to a fake portal, then phoned them. Retool says the caller claimed to be from the IT team and “deepfaked our employee’s actual voice”. The employee handed over an MFA code. Retool notified 27 cloud customers of unauthorised access [ Retool 2023 ] .

In May 2025 the FBI warned of a campaign, running since April that year, using AI-generated voice messages that impersonate senior US officials. The FBI advises treating such messages as unverified, and notes that AI voices can be hard to tell from a known contact [ FBI PSA 2025 ] . Video is no safer. In 2024 Arup confirmed that an employee in Hong Kong sent HK$200 million after a video call in which the “CFO” and colleagues were deepfakes [ Fortune 2024 ] .

Phones also beat email on success rate. In Verizon’s simulation data, the median click rate for email phishing campaigns is 1.4%. For phone-centric methods, voice and text, it is closer to 2%, a 40% increase. Verizon adds that 41% of social engineering breaches involve vectors other than email alone [ Verizon DBIR 2026 ] .

07 How to stop it

You stop pretexting by taking judgement out of the high-risk moments. A helpdesk agent should never have to decide, mid-call, whether a voice sounds genuine. The rule decides.

Then give staff a script. The hardest part of a pretext call is saying no to someone senior who sounds upset. A written line removes the social cost: “I can’t do that on this call. I’ll call you back on the number we have on file.” Make it clear, from the top, that nobody is ever criticised for using it, including on a call that turns out to be the real CEO.

Include your outsourced IT provider. Caesars and Clorox both point to a third-party help desk [ Caesars 8-K ] [ The Record 2025 ] . Ask your provider to show you their written reset procedure, and write the verification steps into the contract.

Policies on paper tell you what should happen. A test call tells you what does. Our penetration testing services can include authorised pretext calls to your helpdesk, finance team and reception, alongside the technical testing, so you see which story got through, which rule broke and what it reached. For a full attack from first call to domain admin, read how we approach red teaming and adversary simulation, and what a social engineering penetration test should hand you. You learn which of your people can be talked round, before an attacker finds out.