A fake IT text and phone call got a Retool employee's MFA codes in 2023. Google Authenticator sync did the rest: 27 crypto accounts hijacked. What happened.
Social Engineering Examples·8 sources
In August 2023, attackers sent Retool employees a text posing as IT, about a payroll sync for open enrolment, with a link to a fake Okta login. One employee entered a password and a one-time code. Then the attacker phoned them, posing as IT again, and got a second code. That let the attacker add their own device, and because the employee's Google Authenticator synced to the cloud, one account opened many others. The attacker took over 27 crypto customers' accounts. Status: Documented incident; deepfake claim unverified.
Several employees got a text from "IT" saying their payroll system was out of sync and needed fixing for open enrolment. The link looked like Retool's own login: retool.okta.com followed by more text. But the real domain was oauthv2.app. Everything before it was just words the attacker chose.
Nearly everyone ignored it. One person logged in on the fake page and entered a password and a one-time code. The attacker then called that person, posing as IT. Retool says the caller knew the office floor plan, colleagues and internal processes. The employee grew more suspicious as the call went on, but still gave one more code.
Tip: read a link from right to left. The real domain is the last part before the first single slash. Anything to its left can say whatever the attacker wants.
| Step | What happened |
|---|---|
| Text | Fake IT message with a lookalike Okta link (smishing) |
| First code | Entered on the fake login page |
| Call | Fake IT on the phone gets a second code (vishing) |
| Own device | The second code let the attacker add their device to the employee's Okta account |
| Cloud sync | Access to the employee's Google account exposed every code stored in Google Authenticator |
| Spread | VPN, then an internal customer-support tool |
| Takeover | Changed emails and reset passwords on 27 cloud customer accounts, all crypto firms |
Retool told the 27 customers on 29 August, closed all internal sessions and restored the hijacked accounts. Customers running Retool on their own servers weren't affected.
Retool's write-up says the caller "deepfaked our employee's actual voice". That's the only evidence. Retool released no audio and no forensic finding, and its head of engineering declined to explain further when Ars Technica asked. Ars called it "a separate and completely unverified claim" and suggested it might distract from Retool's own role. The insider knowledge, which nobody disputes, is enough on its own to explain a convincing call.
Retool never said money was stolen or named its customers. The loss comes from the customer side. Fortress Trust, a Nevada trust company, said four of its customers were hit through a breached cloud vendor. It first said there was "no loss of funds". Five days later its chief executive told Fortune the loss was $12 million to $15 million in crypto, mostly Bitcoin. CoinDesk identified the vendor as Retool. Neither company has confirmed the link on the record.
Fortress said it repaid most affected clients from its own funds within 48 hours, and Ripple covered one larger client.
See also 0ktapus, a similar Okta-themed smishing campaign, and account takeover.
A fake IT text led one employee to a lookalike Okta page, and a follow-up call got a second code. That let the attacker add their own device and reach internal tools.
Retool says the caller faked an employee's voice, but it gave no evidence, and Ars Technica called the claim completely unverified.
Its cloud sync stored the employee's one-time code secrets in their Google account. Once the attacker had that account, they had the codes too.
Retool gave no figure. Fortress Trust, reported to be a Retool customer, said it lost $12 million to $15 million in crypto.
Social Engineering Examples. “Retool breach (2023): a fake IT text, a phone call and synced MFA codes”. Accessed 3 November 2026. https://socialengineeringexamples.com/retool-smishing-vishing-deepfake-2023
the attackers arrived knowing Retool's office floor plan, coworker names, internal processes and the freshly announced Okta login migration. Retool did not say how that detail was gathered, and no source establishes the method; professional-networking profiles, prior breach data and public company announcements are the usual candidates. Treat the sourcing as unknown and the effect as certain.
Employee-facing exposure (roles, org charts, announcements of internal changes) is very hard to eliminate at enterprise scale. The realistic control assumes the attacker already has it and hardens the process it gets used against, rather than trying to hide it.
the attackers stood up a look-alike Okta login page, complete with an MFA form, on a domain they controlled. Retool published the URL: retool.okta.com.[oauthv2.app]/authorize-client/xxx. The only part that decides where the link goes is oauthv2.app; retool.okta.com is hostname text the attacker chose so the URL would read as Retool's real portal at a glance.
Domain-monitoring and typosquat takedown services catch some look-alike domains, and staff can be taught to read a URL from the right, since the registrable domain is the only part that decides destination. Neither is reliable enough to stand alone, so the real control sits at Stage 4.
a text message impersonating a member of Retool's IT team went to several employees, saying their payroll system was out of sync and that it needed to be synced for open enrollment, which affects healthcare coverage. The timing coincided with the real Okta migration, so the request read as routine.
Mobile threat defense and carrier-level filtering catch some lures, but the durable control is training that treats any unsolicited SMS about account or benefits problems with an embedded login link as suspect, to be verified through a known internal channel rather than the number that texted.
almost all employees ignored the text. One logged into the fake portal and entered a password plus a one-time MFA code.
Phishing-resistant, hardware-backed MFA (FIDO2 security keys or passkeys) removes the one-time code entirely, so there is nothing on the fake portal for the employee to hand over. Retool acknowledged that FIDO2 provides resilience to this attack, and did not say why it was not in use.
the attacker phoned the same employee, claiming to be a member of the IT team, and demonstrated familiarity with the floor plan, coworkers and internal processes. Retool says the caller also deepfaked a real employee's voice, a claim no one has verified. The employee grew more and more suspicious during the call and still provided one additional MFA code.
A strict no-OTP-by-phone policy plus callback verification through a known internal number removes the payoff from a convincing call, whatever the caller sounds like. Voice is not authentication even when it is genuinely the right person's voice, which is why the unresolved deepfake question does not change the control.
that second one-time code let the attacker add their own device to the employee's Okta account, so they could generate valid Okta MFA from then on, and it gave them an active Google Workspace session on the attacker's device.
Alerting on new-device MFA enrollment and gating it behind out-of-band approval would have stopped the attacker's device registration from silently granting durable Okta access.
the employee had enabled Google Authenticator's cloud-sync feature, so control of the Google account exposed the OTP seeds stored there. Retool used OTPs for Google, Okta, its VPN and its own internal Retool instances, and argues that this is what turned its multi-factor authentication into single-factor authentication. The attacker reached the VPN and an internal Retool instance used for customer support.
Do not let OTP seeds sync to personal cloud accounts. At the time Google offered enterprises no central switch to disable Google Authenticator sync, so the practical answer was to move off shared-secret OTPs to FIDO2 or passkeys, or to mandate an authenticator whose sync is off by default and end-to-end encrypted.
using that internal support instance, the attacker ran account-takeover actions, changing user emails and resetting passwords, against 27 of Retool's cloud customers, all cryptocurrency firms, then looked around some of the Retool apps.
Segmenting and monitoring internal admin and customer-support tooling, with least-privilege scoping and anomaly detection on bulk email or password changes, limits how far one compromised session reaches into customer accounts.
Retool's own account stops at account takeover and never states that funds were stolen. The theft is documented on the customer side only: Fortress Trust said four of its customers were affected, and its chief executive put the loss at $12M to $15M in crypto. Retool revoked all internal sessions, locked down the affected accounts, restored them to their original state and notified the 27 customers on August 29.
Fast session revocation and account restoration limited Retool's own blast radius. On the customer side, withdrawal holds, address allowlisting and out-of-band confirmation for high-value crypto transactions are the last line once account takeover has already happened. Retool noted that customers who had built their apps with their own threat model in mind repelled the attack even after their accounts were taken over.
Browse by what this case has in common with others in the library.
Fake finance "directors" in WhatsApp groups steered 18+ US investors to NanoBit, a crypto platform that never traded. The SEC…
SABRIC recorded R2.4bn in South African digital banking fraud claims in 2025, driven by phishing, fake bank calls and SIM…
A fraudster spoofed Wells Fargo's real 800 number nine minutes after a legitimate advisor call, phished a 2FA code from…
In 2015, fake emails from Ubiquiti's CEO and a real London lawyer got its finance chief to wire $46.7M. How…
A job candidate at Vidoc Security Lab hid behind a real-time face filter and refused to cover his face. What…
A Chinese national entered Mar-a-Lago after a pool-visit pretext and staff's unverified assumption that she was related to a member.…
In 2022, 0ktapus texted staff fake IT alerts linking to Okta lookalikes, stealing 9,931 logins and MFA codes. Twilio was…
DOJ's first-of-its-kind nationwide takedown charged 61 individuals and entities, including five Ahmedabad, India call-center conglomerates, for an alleged 2012-2016 IRS/USCIS…
US prosecutors charged 28 people over an alleged $65M scheme where callers faked refunds to trick seniors into sending cash.…
Russian GRU-linked Forest Blizzard (APT28/Fancy Bear) used OpenAI's GPT-4-family LLM services to research satellite communication protocols and radar.
Two unencrypted TD Bank backup tapes carrying data on 260,000 customers vanished in transit between Massachusetts offices in 2012.
A scammer spoofed a New Zealand bank's number and obtained authentication codes by claiming they would suspend banking and cancel…
A Chicago hairstylist lost $20,000 to a Phantom Hacker scam. A caller claiming to be Bank of America said a…
A tech-support operator used fake Microsoft/Apple security pop-ups to scare consumers, often older Americans, into calling an India-based support line…
A Lapsus$ affiliate used a contractor's stolen Uber password, likely bought on the dark web, flooded them with MFA push…
Dubai Police raided nine pig butchering centres and arrested 275, and San Diego charged six alleged scam-compound managers and recruiters.…
California's Attorney General and six county DAs found more than 10,000 paper patient records and hazardous/medical waste in unsecured.
SCE reported $131,464 collected by scammers across its 2025 customer fraud complaints. Its warnings describe immediate-disconnection and meter-removal threats as…
A phishing attack resulted in disclosure of 2015 W-2 data for several thousand US employees, exposing SSNs and earnings to…
An Arup employee in Hong Kong sent HK$200M (US$25.6M) after a video call where everyone else was a deepfake. How…
Someone used the White House chief of staff's phone contacts and an AI-like voice to call senators and executives, asking…
In the PG&E scam, a caller claiming to be from the utility threatens to cut the power today. Then a…
An attacker sent a LastPass employee WhatsApp calls and a voicemail using an AI copy of the CEO's voice. The…
The FBI's December 2024 warning on AI-enabled fraud: AI text, fake IDs, cloned voices and deepfake video, and what the…