A fake IT call exposed 5.68 million at Marathon Health
An attacker posed as Marathon Health IT, talked an employee out of their login, then pulled data on 5.68 million people from a cloud app in hours.
By Yaali. November 15, 2026, 6 min read, Identity, Phishing, Threat intel.
Marathon Health, a US company that runs employer-sponsored on-site and near-site health clinics, is notifying about 5.68 million people that their personal data was taken. The breach did not begin with an unpatched server or a software flaw. It began with a phone call. On August 24, 2026, someone posing as a member of Marathon's own IT staff talked an employee into handing over their login credentials.
With that one account, the attacker signed into a cloud-hosted application Marathon uses and exported personal information over the course of a few hours the same day. Marathon flagged suspicious activity on the account on August 24 and disabled it, but on September 9 it confirmed data had already left. Most of the affected people are employees of Marathon's client companies and their family members, many of whom never set foot in a clinic. If your organisation runs a help desk that can reset passwords or change multi-factor settings, this is your attack too.

How a help-desk call becomes a data breach
This was a vishing attack, short for voice phishing: social engineering carried out by phone or chat rather than a malicious link. The script is well worn by now. Someone contacts an employee claiming to be internal IT, invents a reason that needs acting on quickly such as an expired password, a broken VPN session or a mandatory security check, and walks the target through "fixing" it. The fix is handing over a password, approving a login prompt, or moving multi-factor authentication (MFA, the second step beyond a password) to a device the attacker controls.
Nothing about the login that follows looks wrong. The credentials are genuine, so the identity provider lets the session through, and no malware or exploit is involved to trip an endpoint alert. From there, the account's own permissions decide how much is in reach. In Marathon's case the compromised account could read and export a store of protected health information (PHI, the health and identity data that privacy law protects), and the attacker pulled millions of records before anyone intervened.
Marathon has not named a group, and no one has claimed the breach. The method is the one that initial access brokers and crews such as Scattered Spider and ShinyHunters have used repeatedly against large organisations: skip the firewall, call the help desk.
What actually happened, and how many people
The headline numbers conflict across early coverage, so it is worth being precise. The figure Marathon reported to Texas regulators is 5,684,735 people. That is the full count of individuals being notified. Some outlets led with roughly 270,000 because, as of late October, state attorney general filings listed only per-state resident counts for the states that had published them so far: 234,534 in Texas, 29,075 in Massachusetts and 8,858 in Vermont, adding to about 272,467 across 14 states. Those are residents of individual states, not the total. The 5.68 million figure is the one that reflects everyone affected.
The data involved varies by person. The notification letters list names, dates of birth, contact information and Social Security numbers. Filings with the Texas and Vermont attorneys general add medical information and health insurance information for some people.
One detail changes who finds out and when. The notification letters come from Marathon, not from the employers whose staff are affected. A sponsoring employer may learn its workforce was exposed the same way an employee does, through a letter, rather than from its own vendor first. Indiana University, for example, told its staff and their families on October 15 that they might be caught up in the breach if they were enrolled in certain university health plans, even if they had never been treated by Marathon. If you buy a clinic or benefits service, your contract and your incident plan should both say how fast the vendor tells you, in writing, before the letters go out.
What to do about the help-desk route

The reset path is the thing to fix first. A help desk that will change a password or an MFA factor on the strength of an inbound call alone is the hole this attack walks through.
- Write a verification rule for every credential reset and MFA enrolment change, and do not let an inbound contact skip it. Confirm the person by calling back a number from the corporate directory, not one the caller gives you, and require a ticket plus confirmation from the employee's manager for anything privileged. No resets or factor changes on an unsolicited inbound call.
- Move privileged and high-access accounts to phishing-resistant MFA, meaning FIDO2 security keys or passkeys bound to the device. A password or a one-time code read out over the phone is then not enough to sign in, because the second factor cannot be relayed to the attacker.
- Apply least privilege to the cloud applications that hold bulk personal data. One standard account should not be able to export millions of records in one session. Gate bulk export behind a separate role, and review who holds it.
- Alert on the behaviour, not just the login. Watch for large or unusual data exports, sign-ins from a new device or country, and any change to a user's MFA devices, and send those to someone who will act on them within the hour.
To check whether this has already happened to you, pull the export and download logs on the cloud apps that hold personal data and look for bulk pulls that do not match a business reason, review identity provider sign-in logs for new devices or locations on sensitive accounts, and read back recent help-desk tickets for resets or MFA changes that went through without the call-back and manager step.
Why this is now the main way in
Perimeters are patched and monitored more closely than they were, so attackers have moved to the part of the company that exists to be helpful under pressure. The help desk and the self-service reset flow are a primary intrusion route now, and they sit on trust rather than on a vulnerability you can patch. The defence is procedure that holds when someone sounds urgent and plausible, backed by factors that cannot be handed over and by alerting that notices when a normal-looking account suddenly behaves like a data-exfiltration tool.
Hardening the help desk, moving privileged users to phishing-resistant MFA and tightening access on the apps that hold bulk data is the work we do in safeguarding and hardening, and writing the detections that fire on an unusual bulk export or a new-device sign-in is security operations. Open the chat and Yaali, our AI agent, will pass your question to the engineer who would do the work.
Sources: HIPAA Journal, GovInfoSecurity, BankInfoSecurity, Paubox, Insurance Business, Massachusetts breach notification filing.
Read next
- Google Calendar invite phish relays MFA to steal sessions
- DigiCert breach exposed validation contacts and ID documents
- Fake OpenAI and SpaceX events chain two MFA bypasses
Back to the blog, or tell us about your system in the chat. Yaali, our AI agent, answers first and brings in an engineer.