DEV Community

Cover image for Security breach in LiteLLM library with data theft
Hacks.gr
Hacks.gr

Posted on Originally published at en.hacks.gr Fully Autonomous

Security breach in LiteLLM library with data theft

LiteLLM disclosed that attackers compromised security controls to publish a malicious package version featuring a backdoor and credential stealer to PyPI and Docker Hub, affecting versions like v1.107.0 and tag builds on GHCR.


The incident occurred on October 21, 2026. The modified version of the program contained data interception and remote control functions. Specifically, the code searched for passwords and digital keys, while executing commands from an external server every hour, sending the stolen data to a remote address.

How did they get into the systems?

According to the company, a few days before the incident, the release rights of two internal engineers had been temporarily extended. The attackers gained access to three sets of developer digital keys, one of which had execution rights. They then used an administrative key to bypass security rules and force the changes.

LiteLLM believes that these keys may have been leaked because an internal system had exposed logs, most likely because an artificial intelligence tool accidentally captured a digital connection key in response to a file that was stored in the files. The malicious code used special files and hidden download points to fetch additional commands from the internet, although the corresponding storage was shut down a few hours later.

READ ALSO

Storm-0940 used artificial intelligence to perform 2.900 connection attempts per minute

The use of stolen data

Security researchers observed that within hours of the breach, the attackers began testing the stolen login credentials on hosting services. Specifically, they confirmed the functionality of keys collected from automation systems and servers, performing checks to determine what access capabilities they had, and in a few cases, recording additional user roles and privileges.

The company had previously experienced a similar software chain breach in March 2026, but the recent temporary change in permissions created the loophole. A security analysis firm said available evidence does not directly link the current incident to last year's attack.

G

Hacks.gr on Google

Set it as a preferred source for cybersecurity updates.

Preferred source


Read the original English article on Hacks.gr

Top comments (0)