Cloud Security Office Hours Banner

Okta / LAPSUS$ 2022

Step-by-step kill chain mapped to MITRE ATT&CK Cloud, sourced from official post-mortems and primary technical analyses.

January-March 2022 Critical Identity Provider

Okta / LAPSUS$ - A Subcontractor's Support Engineer → Five Days Inside → 366 Customers Exposed → And Two Months Before Anyone Was Told

Between January 16 and 21, 2022, LAPSUS$ had access to the environment of Sitel, a subcontractor providing customer support services to Okta, via a support engineer's laptop. Okta later assessed that up to 366 customers - around 2.5% of its base - may have been affected. The intrusion itself is a fairly ordinary third-party compromise. What made this a defining incident for the industry was the disclosure. Okta received a summary report from Sitel on March 17. Customers found out on March 22, when LAPSUS$ posted screenshots to Telegram, and Okta's security team correlated them to the January incident within ninety minutes of seeing them. Okta received the complete forensic report from Sitel later that same day. For two months, the organizations whose identity provider had been touched did not know. This chain is here for the disclosure failure at least as much as the technical one.

5 days of access: January 16-21, 2022
2 months before customers were told - and only because LAPSUS$ posted
366 customers potentially affected, ~2.5% of Okta's base
Fourth party: a subcontractor of a support provider to the IdP
📄 Okta Security - Official statement on LAPSUS$ claims ↗ 📄 Microsoft - DEV-0537 (LAPSUS$) targeting organizations ↗
Initial Access - Not Okta, and Not Even Okta's Vendor's Employee
01
LAPSUS$ compromised the laptop of a customer support engineer working for Sitel, a subcontractor providing support services to Okta
T1199 - Trusted Relationship T1078 - Valid Accounts

Count the hops. An Okta customer trusts Okta. Okta outsources customer support to Sitel. Sitel staff work on machines Sitel manages. LAPSUS$ compromised one of those machines. So the organizations ultimately exposed were three relationships away from the compromised endpoint, and had no contractual or practical visibility into it - most would not have known Sitel existed. LAPSUS$ was known for buying credentials and recruiting insiders at outsourcers and telecoms precisely because that layer combines broad access with weaker controls and lower scrutiny than the brand-name company it serves.

Chain: Okta customer → Okta → Sitel (support subcontractor) → a support engineer's laptop
Window: January 16-21, 2022 - five days
Actor: LAPSUS$ / DEV-0537, known for recruiting insiders and buying access at outsourcers
Customer visibility into Sitel: Effectively none
Why this layer: Broad access, weaker controls, less scrutiny than the brand it serves
LAPSUS$SitelFourth PartyT1199
🛠 The Access - Support Tooling Is Privileged Tooling
02
Support engineers hold the ability to reset passwords and multi-factor authentication for customer users - limited by design, but powerful by nature
T1098 - Account Manipulation

Okta was clear that support personnel cannot create or delete users, download customer databases, or access source code. But the access they do have is exactly the access an attacker wants from an identity provider: the ability to reset a password, and the ability to reset multi-factor authentication for a user. That is the same capability Scattered Spider obtained by phoning a help desk at MGM eighteen months later, and it is why a support console at an IdP should be treated as a privileged administrative system rather than a customer service tool. The screenshots LAPSUS$ published showed internal applications and a Slack channel, which is also a reminder that support tooling tends to sit alongside internal chat where more context is casually available.

Support cannot: Create or delete users, download customer databases, access source code
Support can: Reset passwords and reset MFA for customer users
Why that matters: It is the precise capability an attacker wants from an IdP
Same capability abused in: Scattered Spider / MGM, via a phone call, eighteen months later
Also exposed: Internal applications and Slack context visible in the published screenshots
Support ConsoleMFA ResetPrivileged ToolingT1098
🕰 The Failure - Two Months of Silence
03
Sitel's summary report reached Okta on March 17; customers learned on March 22 from LAPSUS$ screenshots, and Okta correlated them to the January incident in about ninety minutes

The published timeline is unusually precise and unusually damning. LAPSUS$ posted screenshots at 03:30 UTC on March 22. By 05:00 Okta Security had determined they related to the January Sitel incident. Okta received the complete forensic report from Sitel at 12:27 that same day. So the correlation took ninety minutes once the evidence was public, while the two months before it were spent waiting on a subcontractor's investigation. The customers who might have needed to check their own logs for anomalous password or MFA resets in January were not asked to, because nobody told them. Okta's chief security officer subsequently acknowledged the delay was a mistake. The general lesson is that a contractual notification obligation which permits a two-month forensic process is not a security control - your customers' ability to respond decays to nothing long before the report is finished.

January 16-21: Threat actor active in the Sitel environment
March 17: Okta receives a summary report from Sitel
March 22, 03:30: LAPSUS$ publishes screenshots
March 22, 05:00: Okta correlates them to the January incident - ninety minutes
March 22, 12:27: Okta receives the complete Sitel report
Net effect: Customers could not check their own logs during the window it mattered
Disclosure DelayNotification FailureVendor Forensics
📊 Impact - 366 Customers, and a Lasting Question About IdP Trust
04
Okta assessed that up to 366 customers - roughly 2.5% of its base - may have been affected, and the incident permanently changed how organizations reason about identity provider concentration
T1213 - Data from Information Repositories

The technical blast radius was bounded and Okta's eventual analysis was thorough. The reputational and strategic effect was larger, because it forced a question the industry had been avoiding: an identity provider is a single point of trust for every application behind it, so its supply chain is your supply chain, and its subcontractors are your subcontractors. Okta was breached again through its support organization in October 2023 by a different route, which is why these two chains sit on this page separately - the recurrence is the point. LAPSUS$ itself was disrupted later in 2022 following arrests in the UK.

Potentially affected: Up to 366 customers, ~2.5% of Okta's base
Technical scope: Bounded by what support tooling permits
Strategic effect: Forced the question of identity provider concentration risk
Recurrence: Okta's support organization was compromised again in October 2023, by another route
Actor outcome: LAPSUS$ disrupted later in 2022 following arrests
366 CustomersIdP ConcentrationRepeat IncidentT1213

🛡 How to Defend Against This Chain

Write notification timelines into contracts in hours, not "promptly". This is the concrete, transferable fix. Require vendors to notify you of suspected compromise affecting your data within a defined short window - and specifically require notification on suspicion, before forensics conclude, because the two-month gap here was spent waiting for a complete report. Extend the requirement explicitly to their subcontractors.
Ask your identity provider who else can reset your users' MFA. Support staff, at the vendor and at its outsourcers, hold that capability. Find out who they are, what controls sit around that console, whether actions are logged in a way you can consume, and whether you can restrict or approve support access to your tenant. Several IdPs now offer exactly that control because of this incident.
Monitor your own tenant for the actions a compromised support console would produce. You cannot see inside the vendor, but you can see the outcome: password resets, MFA re-enrolments, and privilege changes on your users. Alert on those as security events rather than help-desk noise, retain the logs yourself, and you retain the ability to answer the question even when nobody tells you to ask it.
Map the fourth party, at least for critical vendors. The exposed endpoint belonged to a subcontractor of a support provider. For the handful of vendors that could hurt you most, ask who they outsource to, what access those parties hold, and how they are assessed. You will not map the whole chain, but for an identity provider it is worth knowing one level further than the contract.
Plan for the incident where the vendor is not the one telling you. Here the notification was effectively a Telegram post. Decide in advance how you validate a public claim about a vendor, who owns the decision to act on unconfirmed reporting, and what defensive steps you take pre-emptively - because waiting for confirmation cost these customers two months.

Related defense topics