A tampered update harvested credentials. What could those credentials reach?
In March, a company with roughly five million experts on its platform installed an update to a routine open source tool. The update had been tampered with. It was built to do one thing: harvest credentials from every system that installs it.
Four months later it was reported in talks at a twenty billion dollar valuation. No perimeter was breached. No zero day was triggered. Someone collected logins and then used them in ways to match the behaviors of the people they stole from.
What the company confirmed
They published their own findings rather than waiting to be reported on, which is worth noting in itself. The update was compromised versions of LiteLLM, installed during the relevant window, designed to walk in and grab every credential it could find from any system that ran them. The size; a limited subset of expert data affected out of a network of nearly five million. No employee data in the mix. Forensics and law enforcement engaged, affected individuals notified in late June with identity protection provided.
Their remediation list showed the map:
- Credential rotation across cloud platforms and systems.
- Restrictive cloud security policies.
- Network controls.
- Twenty four seven detection and response.
Almost every item is about identity and access, not about the compromised package. The package was the delivery mechanism. The credentials were the actual breach.
What it cost
Data work with the company was paused after the incident. A class action followed.
That is not a remediation cost or a notification cost. That is a customer switching off revenue because of what a set of credentials could reach. For a business whose largest accounts are a handful of frontier labs, a single account pausing is the most expensive possible outcome, and it happened without a single system being destroyed.
The part that generalizes
Every company on earth is running dependencies it did not write, on systems holding credentials it has not inventoried, reaching data it has not valued.
Most organizations can tell you which packages they run. Many can tell you which ones have known vulnerabilities. Very few can answer the question that actually decides the size of the incident:
If this dependency harvested every credential on the host it runs on, what could those credentials reach, and what is that worth?
That is not a software composition question, it is an identity question wearing a supply chain costume. The blast radius was never the package, it was the standing access sitting on the machine the package landed on.
Three questions worth putting to your board this quarter
- Which of our systems hold credentials that reach production or customer data, and when were they last rotated?
- If a dependency on any one of those hosts harvested everything it could see, what is the largest thing the attacker reaches?
- What is that worth to us in dollars, and who signed off on that number?
If the third question has no answer, the first two are interesting rather than useful. A board cannot govern an exposure it cannot price.
Next week: what happens to the number when an identity is deleted versus when it is merely hardened. They are not the same, and the difference is larger than most teams expect.
The Identity Risk Play-by-Play. One real identity incident a week: how a login opened the door, what it reached, and what the exposure was worth. Also published at artemion.io/newsletter.