Leaked LiteLLM master key led to malicious package release

2026-10-25 · Cybersecurity

LiteLLM disclosed that an Agent Platform master key accidentally committed to the public LiteLLM-Labs/lite-harness repository in late May 2026 allowed an attacker to push a malicious version of the widely used Python package. SecurityWeek reported the intruders used compromised GitHub CLI OAuth tokens to publish and approve malicious packages on PyPI and Docker, turning the incident into a software supply chain attack. The compromise matters because LiteLLM is a popular library for routing requests to large language models, and a poisoned release can reach downstream applications automatically. Users of the package are advised to check versions and rotate credentials exposed to the leaked key and tokens.

Why it matters：It shows how a single leaked secret in an AI tooling project can cascade into a supply chain compromise affecting many downstream developers, reinforcing the need for secret management and release-pipeline integrity across the AI software stack.

Key points

• The master key was committed to a public GitHub repository in late May 2026.
• GitHub CLI OAuth tokens were used to publish and approve malicious PyPI and Docker packages.
• LiteLLM is a popular Python library for routing LLM requests.

Sources (original reporting)

• The Hacker News
• SecurityWeek

More stories from this day

• Brotli decoder flaw enables RCE in Node.js, Bun and Python
• CISA flags exploited Collabora and Arista VCO flaws
• BlackX UEFI bootkit hits Central Asian diplomats, bypasses Secure Boot
• FCC bans two devices with HiSilicon surveillance chips
• Critical PHP bugs expose web's fragility as AI speeds up exploitation
• Credit markets spooked by 'total meltdown' in riskiest debt; metals traders post record profits